Tensor.Art Green

Unlocking Creativity: The Power of Prompt Words in WritingWriting can sometimes feel tough, especially when you’re staring at a blank page. If you’re struggling to find inspiration, prompt words can be a helpful tool. These words can spark ideas and make writing easier and more fun. Let’s explore how prompt words can boost your creativity and how to use them effectively.What Are Prompt Words?Prompt words are specific words or phrases that inspire you to write. They can be anything from a single word to a short phrase that gets your imagination going. For example, words like "adventure," "friendship," or "mystery" can lead to exciting stories or poems.Why Use Prompt Words?1. Overcome Writer’s Block: If you’re stuck and don’t know what to write, a prompt word can give you a direction to start.2. Spark Creativity: One word can trigger a flood of ideas. It helps you think outside the box.3. Try New Styles: Prompt words encourage you to write in different genres or styles you might not normally explore.4. Build a Writing Habit: Using prompt words regularly can help you develop a consistent writing routine.How to Use Prompt Words1. Make a ListStart by writing down some prompt words that inspire you. Here are a few examples:- Adventure- Dream- Secret- Journey- Change2. Quick Writing ExercisePick a prompt word and set a timer for 10 minutes. Write anything that comes to mind without worrying about making it perfect. This helps you get your ideas flowing.3. Write a Story or SceneChoose a prompt word and try to write a short story or scene based on it. For example, if your word is "mystery," think about a detective solving a case.4. Create a PoemUse a prompt word to write a poem. Let the word guide your ideas and feelings. You can write a simple haiku or free verse.5. Share with FriendsShare your prompt words with friends and challenge each other to write something based on the same word. This can lead to fun discussions and new ideas.Tips for Using Prompt Words- Write Daily: Spend a few minutes each day writing with a prompt word. This builds your skills and keeps your creativity flowing.- Make a Prompt Jar: Write different prompt words on slips of paper and put them in a jar. Whenever you need inspiration, pull one out and start writing.- Reflect on Your Work: After you write, take a moment to think about what you created. What did you like? What can you improve?- Explore Different Genres: Use prompt words to try writing in genres you don’t usually write in, like fantasy or poetry. This helps you grow as a writer. ConclusionPrompt words are a simple yet powerful way to boost your creativity and make writing enjoyable. They can help you overcome blocks, spark new ideas, and develop a consistent writing habit. So, the next time you feel stuck, remember that a single word can lead to amazing stories. Embrace the power of prompt words and watch your creativity soar!

1. Load CLIP VisonDecode the image to form descriptions (prompts), and then convert them into conditional inputs for the sampler. Based on the decoded descriptions (prompts), generate new similar images. Multiple nodes can be used together. Suitable for transforming concepts, abstract things, used in combination with Clip Vision Encode.2. Load CLIPThe Load CLIP node can be used to load a specific CLIP model, CLIP models are used to encode text prompts that guide the diffusion process.*Conditional diffusion models are trained using a specific CLIP model, using a different model than the one which it was trained with is unlikely to result in good images. The Load Checkpoint node automatically loads the correct CLIP model.3. unCLIP Checkpoint LoaderThe unCLIP Checkpoint Loader node can be used to load a diffusion model specifically made to work with unCLIP. unCLIP Diffusion models are used to denoise latents conditioned not only on the provided text prompt, but also on provided images. This node will also provide the appropriate VAE and CLIP and CLIP vision models.*even though this node can be used to load all diffusion models, not all diffusion models are compatible with unCLIP.4. load controlnet modelThe Load ControlNet Model node can be used to load a ControlNet model, Used in conjunction with Apply ControlNet.5. Load LoRA6. Load VAE7. Load Upscale Model8. Load Checkpoint9. Load Style ModelThe Load Style Model node can be used to load a Style model. Style models can be used to provide a diffusion model a visual hint as to what kind of style the denoised latent should be in.* Only T2IAdaptor style models are currently supported.10. Hypernetwork LoaderThe Hypernetwork Loader node can be used to load a hypernetwork. Similar to LoRAs, they are used to modify the diffusion model, to alter the way in which latents are denoised. Typical use-cases include adding to the model the ability to generate in certain styles, or better generate certain subjects or actions. One can even chain multiple hypernetworks together to further modify the model.

Consensus is that the latest generation of Pony SDXL models no linger require "score_9 score_8 score_7" written in the prompt to "look good".//----//It is possible to visualize our actual input to the SD model for CLIP_L ( a 1x768 tensor) as a 16x16 grid , each with RGB values since 16 x 16 x 3 = 768I'll assume CLIP_G in the SDXL model can be ignored. Its assumed CLIP_G is functionally the same but for 1024 dimension instead of 768.So the here we have the prompt : "score_9 score_8_up score_8_up"Then I can do the same but for the prompt : "score_9 score_8_up score_8_up" + XWhere X is some random extremely sus prompt I fetch from my gallery. Assume it to fill up to the full 77 tokens (I set truncate=True on the tokenizer so it just caps off past the 77 token limit)Examples:etc. etc.Granted , first three tokens in the prompt for the 768 encoding greatly influnces the "theme" of the output.But from above images one can see that the "appearance" of the text encoding can vary a lot.Thus , the "best" way to write a prompt is rarely universal.Here I'm running some random text I write myself to check similarity to our "score prompt" (top result should be 100% , so I might have some rounding error) :score_6 score_7_up score_8_up : 98.03% score 8578 : 85.42% highscore : 82.87% beautiful : 77.09% score boobs score : 73.16% SCORE : 80.1% score score score : 83.87% score 1 score 2 score 3 : 87.64% score : 80.1% score up score : 88.45% score 123 score down : 84.62%So even though the model is trained for "score_6 score_7_up score_8_up"we can be kinda loose in how we want to phrase it , if we want to phrase it.Same principle applies for all LoRA and their activation keywords.Negatives are special. The text we write in the negatives are split by whitespace , and the chunks are encoded individually.Link to Notebook if you want to run your own tests:https://huggingface.co/datasets/codeShare/fusion-t2i-generator-data/blob/main/Google%20Colab%20Jupyter%20Notebooks/fusion_t2i_CLIP_interrogator.ipynbI use this thing to search up prompt words using the CLIP_L model//---//These are the most similiar items to the Pony model "score prompt" within my text corpusItems of zero similarity (perpendicular) negative similarity (vector at opposite direction) to encoding are omitted from these results.Note that this are encodings similiar to the "score prompt" trigger encoding , not analysis of what the Pony Model considers good quality.Prompt phrases among my text corpus most similiar to "score_9 score_8_up score_8_up" according to CLIP (the peak of the graph above): Community: sfa_polyfic - 68.3 % holding blood ephemeral dream - 68.3 % Excell - 68.3 % supacrikeydave - 68.3 % Score | Matthew Caruso - 67.8 % freckles on face and body HeadpatPOV - 67.8 % Kazuno Sarah/Kunikida Hanamaru - 67.8 % iers-kraken lun - 67.8 % blob whichever blanchett - 67.6 % Gideon Royal - 67.6 % Antok/Lotor/Regris (Voltron) - 67.6 % Pauldron - 66.7 % nsfw blush Raven - 66.7 % Episode: s08e09 Enemies Domestic - 66.7 % John Steinbeck/Tanizaki Junichirou (Bungou Stray Dogs) - 66.7 % populism probiotics airspace shifter - 65.4 % Sole Survivor & X6-88 - 65.4 % Corgi BB-8 (Star Wars) - 65.4 % Quatre Raberba Winner/Undisclosed - 65.2 % resembling a miniature fireworks display with a green haze. Precision Shoot - 65.2 % bracelet grey skin - 65.2 % Reborn/Doctor Shamal (Katekyou Hitman Reborn!)/Original Male Character(s) - 65.2 % James/Madison Li - 65.1 % Feral Mumintrollet | Moomintroll - 65.1 % wafc ccu linkin - 65.1 % Christopher Mills - 65.0 % at Overcast - 65.0 % Kairi & Naminé (Kingdom Hearts) - 65.0 % with magical symbols glowing in the air around her. The atmosphere is charged with magic Ghost white short kimono - 65.0 % The ice age is coming - 65.0 % Jonathan Reid & Bigby Wolf - 65.0 % blue doe eyes cortical column - 65.0 % Leshawna/Harold Norbert Cheever Doris McGrady V - 65.0 % foxtv matchups panna - 65.0 % Din Djarin & Migs Mayfeld & Grogu | Baby Yoda - 65.0 % Epilogue jumps ahead - 65.0 % nico sensopi - 64.8 % 秦风 - Character - 64.8 % Caradoc Dearborn - 64.8 % caribbean island processing highly detailed by wlop - 64.8 % Tim Drake's Parents - 64.7 % probiotics hardworkpaysoff onstorm allez - 64.7 % Corpul | Coirpre - 64.7 % Cantar de Flor y Espinas (Web Series) - 64.7 % populist dialog biographical - 64.7 % uf!papyrus/reader - 64.7 % Imrah of Legann & Roald II of Conte - 64.6 % d brown legwear - 64.6 % Urey Rockbell - 64.6 % bass_clef - 64.6 % Royal Links AU - 64.6 % sunlight glinting off metal ghost town - 64.6 % Cross Marian/Undisclosed - 64.6 % ccu monoxide thcentury - 64.5 % Dimitri Alexandre Blaiddyd & Summoner | Eclat | Kiran - 64.5 %

TensorArt で最新バージョンの Stable Diffusion を体験してボーナスを獲得しましょう!SD3.5L のオンライン生成とモデル トレーニングがすべて TensorArt でサポートされるようになりました 🥳全部させるには テンソル人 SD3.5Lの楽しさを体験していただくために、 期間限定40％オフ SD3.5 トレーニングの割引と、 サプライズ報酬 😉楽しい準備をしましょう！ ✨SD3.5Lモデルで生成した画像を投稿してください50クレジットタグを含める必要があります: SD3.5LSD3.5Lモデル（CKPT、LoRAなど）の公開300クレジットSD3.5Lモデルを使用したAIツールを公開300クレジットラッキーフロアプレゼント！ ✨35 がこのイベントのラッキーナンバーになりました🙋N 番目に公開されたモデルと AI ツールには、 35ドルのボーナス ！ 「有効機種一覧」 Google ドキュメントで利用可能になり、毎週平日午前 10 時 (UTC) に更新されます。 勝利のチャンスを楽しみにしていてください!👉 SD3.5 List of Valid Models👀N番目に公開されたモデルがカウントされることに注意してください 10月28日00:00 UTCから開始。以前に発売されたモデルは並び替えの対象外となります。 イベント日10月28日00:00 ～ 11月17日23:59 UTCルール準拠していない提出物は失格となり、次の有効なモデルに移行されます。モデルはオリジナルである必要があります。再版や統合は許可されていません。NSFW コンテンツ、児童搾取、有名人の肖像、または低品質なものは含まれません。各タイプの報酬とボーナスは 1 回のみ受け取ることができます。ユーザーのアバターとニックネームをシステムのデフォルトにすることはできません。

Simple comparison of the models (Base on Personal opinion)1. SDXL: Best for producing high-quality, realistic images and works well with various styles. It excels in detail enhancements, especially for faces, and offers many good LoRA variations. It generates large, sharp images that are perfect for detailed projects. However, some images may appear distinctly "AI-generated," which might not suit everyone's preference.2. Pony Diffusion: Known for its artistic flexibility, it doesn’t copy specific artist styles but gives beautiful, customizable results. It is also fine-tuning capable, producing stunning SFW and NSFW visuals with simple prompts. Users can describe characters specifically, making it versatile for various creative needs.3. SD3: Focuses on generating realistic and detailed images, offering more control and customization than earlier versions. Despite the many controversies surrounding SD3, SD3 is also widely used in Comfyui.4. Flux: Ideal for fixing image issues like anatomy or structure problems. It enhances image quality by adding fidelity and detail, particularly in text and small image elements, can provide a clearer concept, better prompt implementation with more natural depiction. 5. Kolors: Great for styling, and make colorful and vibrant artwork, especially in fantasy or creative designs.6. Auraflow: Specializes in smooth, flowing images, often with glowing or ethereal effects, perfect for fantasy or sci-fi themes.And if you want to combine the best of different AI models? you can try my workflow or my ai tool:SDXL MergeSimple - this simple workflow can merge 2 checkpoints with the same base, Pony + FLUX Fixer - and you can try this ai tool if you want to merging 2 different base, since FLUX good at fixing image, text, and small detail, so it will be effective without having to work twice.Finally, all of this is my personal opinion from what I experienced, How about you? do you have a different opinion? and which model do you prefer? share your thoughts in the comments below! let's open the discussion!

Full article can be found here : Stable Diffusion 3.5 Large Fine-tuning TutorialImages should be cropped into these aspect ratios:If you need help automatically pre-cropping your images, this is a lightweight, barebones [script](https://github.com/kasukanra/autogen_local_LLM/blob/main/detect_utils.py) I wrote to do it. It will find the best crop depending on:(1024, 1024), (1152, 896), (896, 1152), (1216, 832),(832, 1216), (1344, 768), (768, 1344), (1472, 704)1. Is there a human face in the image? If so, we’ll do the cropping oriented around that region of the image.2. If there is no human face detected, we’ll do the cropping using a saliency map, which will detect the most interesting region of the image. Then, a best crop will be extracted centered around that region.Here are some examples of what my captions look like:k4s4, a close up portrait view of a young man with green eyes and short dark hair, looking at the viewer with a slight smile, visible ears, wearing a dark jacket, hair bangs, a green and orange background k4s4, a rear view of a woman wearing a red hood and faded skirt holding a staff in each hand and steering a small boat with small white wings and large white sail towards a city with tall structures, blue sky with white clouds, cropped If you don't have your own fine-tuning dataset, feel free to use this dataset of paintings by John Singer Sargent (downloaded from WikiArt and auto-captioned) or a synthetic pixel art dataset.I’ll be showing results from several fine-tuned LoRA models of varying dataset size to show that the settings I chose generalize well enough to be a good starting point for fine-tuning LoRA.repeats duplicates your images (and optionally rotates, changes the hue/saturation, etc.) and captions as well to help generalize the style into the model and prevent overfitting. While SimpleTuner supports caption dropout (randomly dropping captions a specified percentage of the time), it doesn’t support shuffling tokens (tokens are kind of like words in the caption) as of this moment, but you can simulate the behavior of kohya’s sd-scripts where you can shuffle tokenswhile keeping an n amount of tokens in the beginning positions. Doing so helps the model not get too fixated on extraneous tokens.Steps calculationMax training steps can be calculated based on a simple mathematical equation (for a single concept):There are four variables here:Batch size: The number of samples processed in one iteration.Number of samples: Total number of samples in your dataset.Number of repeats: How many times you repeat the dataset within one epoch.Epochs: The number of times the entire dataset is processed.There are 476 images in the fantasy art dataset. Add on top of the 5 repeats from multidatabackend.json . I chose a train_batch_size of 6 for two reasons:This value would let me see the progress bar update every second or two.It’s large enough in that it can take 6 samples in one iteration, making sure that there is more generalization during the training process.If I wanted 30 or something epochs, then the final calculation would be this:represents the number of steps per epoch, which is 396.As such, I rounded these values up to 400 for CHECKPOINTING_STEPS .⚠️ Although I calculated 11,900 for MAX_NUM_STEPS, I set it to 24,000 in the end. I wanted to see more of samples of the LoRA training. Thus, anything after the original 11,900 would give me a good gauge on whether I was overtraining or not. So, I just doubled the total steps 11,900 x 2 = 23,800, then rounded up.CHECKPOINTING_STEPS represents how often you want to save a model checkpoint. Setting it to 400 is pretty close to one epoch for me, so that seemed fine.CHECKPOINTING_LIMIT is how many checkpoints you want to save before overwriting the earlier ones. In my case, I wanted to keep all of the checkpoints, so I set the limit to a high number like 60.Multiple conceptsThe above example is trained on a single concept with one unifying trigger word at the beginning: k4s4. However, if your dataset has multiple concepts/trigger words, then your step calculation could be something like this so:2 concepts [a, b]Lastly, for learning rate, I set it to 1.5e-3 as any higher would cause the gradient to explode like so:The other relevant settings are related to LoRA.{ "--lora_rank": 768, "--lora_alpha": 768, "--lora_type": "standard" } Personally, I received very satisfactory results using a higher LoRA rank and alpha. You can watch the more recent videos on my YouTube channel for a more precise heuristic breakdown of how image fidelity increases the higher you raise the LoRA rank (in my opinion).Anyway, If you don’t have the VRAM, storage capacity, or time to go so high, you can choose to go with a lower value such as 256 or 128 .As for lora_type , I’m just going with the tried and true standard . There is another option for the lycoris type of LoRA, but it’s still very experimental and not well explored. I have done the deep-dive of lycoris myself, but I haven’t found the correct settings that produces acceptable results.Custom config.json miscellaneousThere are some extra settings that you can change for quality of life.{ "--validation_prompt": "k4s4, a waist up view of a beautiful blonde woman, green eyes", "--validation_guidance": 7.5, "--validation_steps": 200, "--validation_num_inference_steps": 30, "--validation_negative_prompt": "blurry, cropped, ugly", "--validation_seed": 42, "--lr_scheduler": "cosine", "--lr_warmup_steps": 2400, } "--validation_prompt": "k4s4, a waist up view of a beautiful blonde woman, green eyes""--validation_guidance": 7.5 "--validation_steps": 200 "--validation_num_inference_steps": 30 "--validation_negative_prompt": "blurry, cropped, ugly""--lr_scheduler": "cosine""--lr_warmup_steps": 2400These are pretty self-explanatory:"--validation_prompt"The prompt that you want to use to generate validation images. This is your positive prompt."--validation_negative_prompt"Negative prompt."--validation_guidance"Classifier free guidance (CFG) scale."--validation_num_inference_steps"The number of sampling steps to use."--validation_seed"Seed value when generating validation images."--lr_warmup_steps"SimpleTuner has set the default warm up to 10% of the total training steps behind the scenes if you don’t set it, and that’s a value I use often. So, I hard-coded it in (24,000 * 0.1 = 2,400). Feel free to change this."--validation_steps"The frequency at which you want to generate validation images is set with "--validation_steps". I set mine to 200, which is a 1/2 of 400 (number of steps in an epoch for my fantasy art example dataset). This means that I generate a validation image every 1/2 of an epoch. I suggest generating validation images at least every half epoch as a sanity check. If you don’t, you might not be able to catch errors as quickly as you can.Lastly is "--lr_scheduler" and "--lr_warmup_steps".I went with a cosine scheduler. This is what it will look like:### Memory usageIf you aren’t training the text encoders (we aren’t), `SimpleTuner` saves us about `10.4 GB` of VRAM.With the settings of `batch size` of `6` and a `lora rank/alpha` of `768`, the training consumes about `32 GB` of VRAM.Understandably, this is out of the range of consumer `24 GB` VRAM GPUs. As such, I tried to decrease the memory costs by using a `batch size` of `1` and `lora rank/alpha` of `128` .Tentatively, I was able to bring the VRAM cost down to around `19.65 GB` of VRAM.However, when running inference for the validation prompts, it spikes up to around `23.37 GB` of VRAM.To be safe, you might have to decrease the `lora rank/alpha` even further to `64`. If so, you’ll consume around `18.83 GB` of VRAM during training.During validation inference, it will go up to around `21.50 GB` of VRAM. This seems safe enough.If you do decide to go with the higher spec training of `batch size` of `6` and `lora rank/alpha` of `768` , you can use the `DeepSpeed` config I provided [above](https://www.notion.so/Stable-Diffusion-3-5-Large-Fine-tuning-Tutorial-11a61cdcd1968027a15bdbd7c40be8c6?pvs=21) if your GPU VRAM is insufficient and you have enough CPU RAM.

In the world of artificial intelligence (AI), especially in training image-based models, the terms DORA, LoRA, and LOKR often play different but complementary roles in developing more efficient and accurate AI models. Each has a unique approach to understanding data, adapting models, and involving developers in the process. This article will discuss what DORA, LoRA, and LOKR are in the context of AI image training, as well as their respective strengths and weaknesses.1. DORA (Distributed Organization and Representation Architecture) in AI Image Training DORA is a model better known in the fields of cognitive science and AI, focusing on how systems understand and represent information. Although not commonly used directly in AI image training, DORA's principle of distributed representation can be applied to how models understand relationships between elements in an image—such as color, texture, shape, or objects—and how those elements are connected in a broader context.Strengths: Understanding complex relationships: DORA allows AI models to understand complex relationships between objects in an image, crucial for tasks such as object recognition or object detection.Strong generalization: Helps models learn more abstract representations from visual data, allowing for object recognition even with variations in form or context.Weaknesses: Less specific for certain visual tasks: DORA may be less optimal for tasks requiring high accuracy in image details, such as image segmentation.Computational complexity: Using a model based on complex representations like DORA requires more computational resources.2. LoRA (Low-Rank Adaptation) in AI Image Training LoRA is a method widely used in AI for fine-tuning large models without requiring significant resources. LoRA reduces model complexity by factoring heavy layers into low-rank representations. This allows for adjustments to large models (such as Vision Transformers or GANs) without retraining the entire model from scratch, saving time and cost.Strengths: Resource efficiency: LoRA enables faster and more efficient adaptation of models, especially when working with large models and smaller datasets.Reduces overfitting: Since only a small portion of the parameters are adjusted, the risk of overfitting is reduced, which is essential when working with limited image datasets.Pretrained model adaptation: LoRA allows for the reuse of large pretrained models trained on vast datasets, making it easier to adapt them to more specific datasets.Weaknesses: Limited to minor adjustments: LoRA is excellent for minor adjustments, but if significant changes are needed or if the dataset differs greatly from the original, the model may still require deeper retraining.Dependent on base model: The best results from LoRA heavily rely on the quality of the pretrained model. If the base model is not strong enough, the adapted results may be unsatisfactory.3. LOKR (Locus of Control and Responsibility) in AI Image Training LOKR, derived from psychology, refers to how a person perceives control and responsibility over something. In the context of AI development, this concept can be applied to how developers feel responsible for and control the training process of the model. Developers with an internal locus of control feel they have full control over the training process, while those with an external locus of control might feel that external factors such as datasets or hardware are more influential.Strengths: Better decision-making: Developers with an internal locus of control are usually more focused on optimizing parameters and trying various approaches to improve results, which can lead to better AI models.High motivation: Developers who feel in control of the training outcomes are more motivated to continuously improve the model and overcome technical challenges.Weaknesses: Challenges with external factors: Developers with an external locus of control might rely too much on external factors such as the quality of the dataset or available hardware, which can limit innovation and control over the training process.Not directly related to AI technicalities: While this concept provides good psychological insights, it does not offer direct solutions in the technical training of AI models.Conclusion DORA, LoRA, and LOKR bring different perspectives to AI image-based training. DORA offers insight into how models can understand complex relationships in images, though it comes with computational challenges. LoRA is highly useful for adapting large models in a more resource-efficient way, but has limitations if larger changes are required. Meanwhile, LOKR, although derived from psychology, can influence how AI developers approach training, especially in terms of control and responsibility. By understanding the strengths and weaknesses of each approach, developers can more effectively choose the method that best fits the specific needs of their AI projects, maximizing both efficiency and model performance in processing images.

Hello everyone.I have been working more intensively with various AI tools in the last few days and weeks. In this article I would like to briefly share my opinion on the "workflows" that you can create with ComfyUI.First of all, my computer is not the "more expensive, faster, better" type. It is a Ryzen 5 with a GForce 3060 Ti. So it is not bad, but by far not the best for training LoRAs, checkpoints or other AI things. It simply takes longer than with a Ryzen 9 and a GForce 4090 ;)But back to ComfyUI and the workflows.Since I have only been working with it for a few days, before that I used A1111 (Stable Diffusion), I am of course far from someone who can give you tips if you have problems. But one thing is certain: ComfyUI is definitely extremely faster than A1111 when creating images.With my current setup, I need over 2 minutes per XL image and almost 5 minutes for FLUX-based images with A1111. Anyone who can do a bit of math knows that this is really incredibly slow...ComfyUI, on the other hand, even with my setup, needs less than 20 seconds for an XL image and almost 60 seconds for a FLUX-based image. Of course, that depends on the workflow.The problem with ComfyUI, in my opinion, is that it is not at all beginner-friendly. There is a "standard" workflow, but that is not enough. After all, we want to integrate or test various checkpoints, LoRAs or other things.So you start and look at the different options... and then... then you don't know what to do next. So without looking at various documentation or examples, you will have an extremely difficult time understanding this tool.If we take the "fresh" installation of ComfyUI, after a long browse you will find that the things you actually want are "not" there. This includes things like using placeholders or a "better" way to save the files you create.This brings us to the possible extensions. Like in so many other communities, there are a huge number here. Unfortunately, this also makes things very confusing. Again, you have to look closely at what you want, need or expect, but even then it doesn't mean that the extension does what you want.The worst thing about ComfyUI in my opinion is the confusing menu and it gets worse with every extension. If you just look at the "Workflow" tool here in Tensort.Art, you immediately understand what I mean.Still. ComfyUI is a very good and powerful tool. Most importantly, it is much faster than the other tools I have tried so far. I also really like the flexibility of the tool. However, it could have a "better" menu to make it more user-friendly.If you haven't done it before: It's worth to check it out.

INSTRUCTIONS:This is a very simple workflow, just upload your image and press RUN.PROMPT basically does not need to be modified, but you can still add more Halloween elements to make the theme richer.Hope you all have a good time.PROMPT:(masterpiece), ((halloween elements)),a person, halloween striped thighhighs, witch hat, grin, (ghost), sweets, candy, candy cane, cookie, string of flags, halloween costume, jack-o'-lantern bucket, halloween, pumpkins,black cat,halloween,little ghost,magic robe,autumn leaves,candle,skull, 3d cg.Negative PROMPT:None.Below is the workflow link:https://tensor.art/workflows/786144487641608308Below is the AI-tool link:https://tensor.art/template/786150277257599620model used:CKPThttps://tensor.art/models/757279507095956705/FLUX.1-dev-fp8

Looking to stand out this Halloween with a fierce, captivating costume? Dive into our 50 Beauty Monster and Creature Inspirations for Halloween 2024!From the alluring vampire queen with fangs and pale skin, to the mystical forest spirit with branches for hair, this list features a variety of iconic, feminine creatures to embody. Each entry provides five key characteristics to make your costume pop with creativity. Whether you want elegance, spookiness, or a combination of both, these ideas will help you slay this Halloween!Vampire: Fangs, cloak, pale skin, red lips, pointed ears.Witch: Pointed hat, broomstick, black dress, potion bottles, striped stockings.Medusa: Snake hair, stony gaze, green skin, gold jewelry, ancient toga.Banshee: Ghostly white dress, flowing hair, haunting scream, pale makeup, chains.Succubus: Bat wings, red dress, horns, glowing eyes, tail.Werewolf: Furry ears, sharp claws, fangs, torn clothes, wild hair.Mermaid: Scales, seashell bra, fishtail, wet-look hair, pearls.Harpy: Feathered wings, talons, bird-like eyes, fierce expression, ragged clothes.Fairy: Sparkling wings, flower crown, wand, glittery makeup, light dress.Zombie: Torn clothes, blood stains, decayed skin, lifeless eyes, open wounds.Siren: Wet-look hair, seashell jewelry, seaweed skirt, alluring voice, eerie glow.Elf: Pointed ears, elegant gown, bow and arrow, long hair, ethereal glow.Gorgon: Snake tail, golden scales, slit eyes, regal crown, sharp claws.Mummy: Wrapped in bandages, dark eye makeup, jewelry, ancient amulet, dusty appearance.Ghost: Flowing white sheet, transparent, eerie wail, glowing eyes, pale hands.Queen of the Dead: Black gown, skull crown, skeletal makeup, dark veil, red roses.Demoness: Red skin, black horns, tail, wings, sharp claws.Bride of Frankenstein: Black and white hair, stitched skin, bride gown, lightning bolts, scars.Voodoo Priestess: Skull face paint, voodoo doll, bones, beads, tribal clothing.Phoenix: Fiery wings, flame patterns, red and orange outfit, glowing skin, feathers.Chimera: Lion mane, snake tail, dragon wings, golden eyes, muscular build.Spider Queen: Black web dress, spider crown, long legs, red eyes, venomous fangs.Lady Death: Black cloak, scythe, skeletal hands, skull mask, dark aura.Nymph: Nature gown, flowers in hair, earthy tones, glowing skin, delicate wings.Selkie: Fur cloak, watery skin, ocean jewels, seal tail, wet hairGiantess: Massive build, oversized clothes, earthy skin, towering presence, big jewelry.Forest Witch: Mossy cloak, animal bones, green skin, potions, tree branches in hair.Dragoness: Scaly skin, horns, tail, fiery breath, armored chestplate.Lilith: Dark wings, black robe, seductive look, glowing red eyes, ancient symbols.Hag: Wrinkled skin, tattered clothes, long nose, hunched posture, warts.Valkyrie: Winged helmet, sword, battle armor, braided hair, shield.Troll Woman: Green skin, sharp tusks, club, fur clothes, wild hair.Ice Queen: Frosted crown, shimmering cape, blue skin, ice staff, glowing cold eyes.Scarecrow: Straw-filled body, stitched mouth, tattered hat, pumpkin head, patched overalls.Djinn: Flowing robes, magic lamp, glowing eyes, ornate jewelry, smoke swirling around.Cheshire Cat: Striped fur, wide grin, cat ears, mischievous eyes, tail.Swamp Creature: Muddy skin, algae hair, webbed fingers, water plants, gills.Basilisk Queen: Reptilian skin, glowing eyes, snake tail, venomous fangs, ancient armor.Lamia: Snake body, golden armor, hypnotic eyes, deadly claws, venomous bite.Wendigo Woman: Deer antlers, skeletal body, glowing eyes, fur cloak, sharp claws.Shadow Witch: Black shadowy figure, dark veil, glowing red eyes, spectral form, floating.Frost Maiden: Icicle crown, snowflake gown, pale blue skin, icy breath, shimmering frost.Baba Yaga: Hunched back, long nose, flying broom, warts, iron teeth.Kitsune: Fox ears, fluffy tail, red kimono, mystical powers, mask.Forest Spirit: Tree branches for hair, bark-like skin, moss gown, glowing eyes, ethereal glow.Plague Doctoress: Black cloak, plague mask, long gloves, eerie eyes, dark potions.Dullahan: Headless woman, flowing black cloak, horse-riding, holding a skull, eerie lantern.Succubus Queen: Leather bodice, wings, horns, glowing eyes, seductive aura.Dryad: Bark skin, leaves in hair, tree branches for arms, glowing green eyes, earthy gown.Banshee Queen: Flowing black dress, ghostly hair, skeletal hands, pale skin, sorrowful wail.settings usedAll created using Juggernaut SDXL modelsteps 25cfg 6dpmpp_2m karrasnot all creature recognize well by the checkpoint, you may use LoRA or other checkpoint if needed to create certain characterWith these 50 beauty monster and creature inspirations, you're all set to embrace the eerie, enchanting side of Halloween 2024. Whether you choose to transform into a seductive vampire, a magical forest spirit, or a chilling banshee queen, each idea is designed to make you stand out in both style and spookiness. Let your creativity soar this Halloween, and enjoy bringing these unique creatures to life. Get ready to slay (literally!) with hauntingly beautiful looks that will leave everyone spellbound!

He actualizado todos mis modelos para que la gente pueda generar imágenes de manera ilimitada y gratuita con ellos, la descarga sigue sujeta al pago del bufet, asi que adelante, den rienda suelta a su creatividad.//I've updated all my models so that people can generate unlimited images with them for free, downloading them is still subject to paying the buffet, so go ahead and unleash your creativity.

You migh have seen this kind of images in the past if you've girly tastes when navigate on pinterest, well guess what? I'll teach you about some parammeters to enhance your Pony SDXL future generations. It's been a while since my last post, today I'll teach you about a cool feature launched by NVIDIA on July 22, 2024. For this task I'll provide an alternative workflow (Diffusion Workflow) for SDXL. Now lets go with the content.ModelsFor my research (AI Tool) I decided to use the next models:Checklpoint model: https://tensor.art/models/757869889005411012/Anime-Confetti-Comrade-Mix-v30.60 LoRA: https://tensor.art/models/7025156632998356040.80 LoRA: https://tensor.art/models/757240925404735859/Sailor-Moon-Vixon's-Anime-Style-Freckledvixon-1.00.75 LoRA: https://tensor.art/models/685518158427095353NodesThe Diffusion Workflow has many nodes I've merged in single nodes I'll explain them below, remember you can group nodes and edit their values to enhance your experience.👑 Super Prompt Styler // Advanced Manager (CLIP G) text_positive_g: positive prompt, subject of the scene (all the elements the scene is meant for, LoRA Keyword activators).(CLIP L) text_positive_l: positive prompt, all the scene itself is meant (composition, lighting, style, scores, ratings).text:negative: negative prompt.◀Style▶: artistic styler, select the direction for your prompt, select 'misc Gothic' for halloween direction.◀Negative Prompt▶: prepares the negative prompt splitting it in two (CLIP G and CLIP L) for the encoder.◀Log Prompt▶: add information to metadata, produces error 1406 when enabled, so turn it off.◀Resolution▶: select the resolution of your generation.👑 Super KSampler // NVIDIA Aligned Stepsbase_seed: similar to esnd (know more here).similarity: this parameter influences base_seed noise to be similar to noise_seed value.noise_seed: the exact same noise seed you know.control after generate: dictates the behavior of noise_seed.cfg: guidance for the prompt, read about <DynamicThresholdingFull> to know the correct value. I recomend 12sampler_name: sampling method.model_type: NVIDIA sampler for SDXL and SD models.steps: the exact same steps you know, dictates how much the sampling denoises the noise injected.denoise: the exact same denoise you know, dictates the strong the sampling denoises the noise injected.latent_offset: select between {-1.00 Darker to 1.00 Brighter} to modify the input latent, any value different than 0 adds information to enhance final result.factor_positive: upscale factor for the conditioning.factor_negative: upscale factor for the conditioning.vae_name: the exact same vae you know, dictates how the noise injected is denoised by the sampler.👑 Super Iterative Upscale // Latent/on Pixel Spacemodel_type: NVIDIA sampler for SDXL and SD models.steps: number of steps the UPSCALER (Pixel KSampler) will use to correct the latent on pixel space while upscaling it.denoise: dictates the strenght of the correction on the latent on pixel space.cfg: guidance for the prompt, read about <DynamicThresholdingFull> to know the correct value. I recomend 12upscale_factor: number of times the upscaler will upscale the latent (must match factor_positive and factor_positive) upscale_steps: dictates the number of steps the UPSCALER (Pixel KSampler) will use to upscale the latent.MiscellaneousDynamicThresholdingFullmimic_scale: 4.5 (Important value. go to learn more)threshold_percentile: 0.98mimic_mode: half cosine downmimic_scale_min: 3.00cfg_mode: half cosine downcfg_scale_min: 0.00sched_val: 3.00separate_feature_channels: enablescaling_starpoint: meanvariability_measure: ADinterpolate_phi: 0.85Learn more: https://www.youtube.com/watch?v=_l0WHqKEKk8Latent OffsetLearn more: https://github.com/spacepxl/ComfyUI-Image-Filters?tab=readme-ov-file#offset-latent-imageAlign Your StepsLearn more: https://research.nvidia.com/labs/toronto-ai/AlignYourSteps/LayerColor: Levelsset black_point = 0 (base level of black)set white_point = 255 (base level of white)Set output_black_point = 20 (makes blacks less blacks)Set output_white_point = 220 (makes whites less whites)Learn more: https://docs.getsalt.ai/md/ComfyUI_LayerStyle/Nodes/LayerColor%3A%20Levels/LayerFilter:Filmcenter_x: 0.50center_y: 0.50saturation: 1.75vignete_intensity: 0.20grain_power: 0.50grain_scale: 1.00grain_sat: 0.00grain_shadows: 0.05grain_highs: 0.00blur_strenght: 0.00blur_focus_spread: 0.1 focal_depth: 1.00Learn more: https://docs.getsalt.ai/md/ComfyUI_LayerStyle/Nodes/LayerFilter%3A%20Film/?h=filmResultAi Tool: https://tensor.art/template/785834262153721417DownloadsPony Diffusion Workflow: https://tensor.art/workflows/785821634949973948

So I set out with what I thought was a simple idea:“Start with an image of someone’s face and turn that into a spooky Halloween character, with costume, makeup and full Facepaint with a spooky background.”BUT it had to look enough like them at the end - that they would be pleased with the result…The starting point was easy - I wanted to train a Halloween LoRA on lots of images of people wearing Halloween Facepaint - so I did that…A couple of the 48 images i used to train with:So I had a Flux LoRA - now I tested that in Tensor.Art with simple “Man in Halloween Facepaint”, “Woman in Halloween Facepaint”So far so good, I thought ok, this is going to be easy peasy!At this point (End of September 2024) there were limited options in TA for Flux Face swap… (No Pulid available then) so I started trying with Facedetailer…I built out the workflow - made a separate flow for the background - and was all excited…But no matter what i tried (and I tried a lot!) the facedetailer would wipe out the Facepaint from the Lora - restoring the face back to the original person, nice and clean, or with a half hearted smear of greasepaint.Or it would look nothing at all like the person and the makeup would look like it was a badly stuck on mask…So i went back to my Discord buddies and we talked about the options - and decided to try Reactor nodes with insightface…It would generate a Florence description of the original reference face (cropped) - build a dummy Halloween Image with a lookielikie from the description and with Facepaint - and then reactor the ref face back over the top (or so i thought)But the Reactor’d one cleaned up the face and removed 90% of the makeup and it didn’t want to do the costume or background at all the way I had envisaged… as soon as I gave it enough freedom to be creative, the reference person was lost completely…I think by now people in all my discord groups were sick of me asking for ideas on how to do this - I tried every setting and balance on reactor nodes.Could I use an llm to rewrite the visual description of the face to include the Halloween description first, and so on.I looked at IPAdapter and using Depth maps - but although they captured the shape of the face - they couldn’t preserve the familiar features through costume stylemakeup.At this point - I pretty much gave up in disgust… I put out a final round of help requests on various discord’s and went onto another projectA few days later my good friend told me “ hey - finally they released Pulid for Flux on TA!” - and I already had built Flux Pulid workflows for face swapping the previous week on my MimicPC Cloud version of Comfyui (where you can load any kind of node and model you want and really design and play with freedom) so I started to regain my enthusiasm…I managed to merge some of the earlier ideas for generating the Halloween style with LLM’s and a Joycaption of the cropped reference face - and the Flux Pulid face swaps - and experimented with the positioning of the LoRA to get maximum effect - and was finally able to release a workflow and AI Tool that did what i had seen in my head those few weeks back when I started… https://tensor.art/template/785795972520313546And the workflow - https://tensor.art/workflows/785793305345589081And the LoRA - https://tensor.art/models/785804669831296337If you have enjoyed my article - please like and use my AI Tools and Models…I welcome comments and constructive feedback.

Halloween is right around the corner, and it’s time to infuse your generation models with a touch of spooky magic! Whether you’re crafting images, stories, or even interactive AI experiences, this guide will help you conjure up the best Halloween-themed content for 2024. Let’s dive into some tips and tricks to make your generative AI creations truly spine-chilling! 🧛‍♂️🕸️1. Theme Selection: Classic Horror vs. Modern ThrillsStart with deciding the tone of your Halloween project. Are you going for classic horror, with haunted houses, creepy forests, and gothic vibes? Or are you leaning towards modern Halloween with neon lights, cyberpunk ghosts, or playful skeletons?Classic horror themes like vampires, witches, and ghosts never go out of style, but blending them with modern elements (think AI-enhanced haunted tech or neon-lit crypts) can bring a fresh twist to your content.2. Prompts and Inspiration IdeasFor image generation, try prompts that capture the Halloween atmosphere:"A haunted Victorian mansion under a full moon, surrounded by fog and dark twisted trees""A neon-lit skeleton playing an electric guitar on a cyberpunk street""A witch stirring a glowing cauldron, with enchanted bats swirling around"For story generation, build a suspenseful atmosphere with prompts like:"On Halloween night, a group of friends discovers a hidden portal in an abandoned amusement park...""A town where every carved pumpkin holds the soul of a spirit seeking freedom"Don't be afraid to add a bit of humor to your Halloween stories, like:"A vampire who’s afraid of the dark trying to overcome his fear"3. Style Adjustments: The Magic of LightingLighting can make or break the eerie ambiance of your Halloween images. Play with shadows, moonlit scenes, or dimly lit rooms to add that sense of unease.Experiment with different color palettes—orange, black, and purple are classics, but consider adding splashes of neon green or eerie blue for a modern twist.For a vintage horror feel, use grainy textures, sepia tones, or black-and-white effects to mimic old horror films.4. Interactive Elements: Make it a Thrilling ExperienceFor those building interactive experiences, consider adding branching storylines where users can explore haunted locations or solve spooky mysteries.Add random elements to make the experience unpredictable—imagine a haunted AI guide that offers different creepy clues each time users interact with it.Build suspense with sound effects like whispering winds, distant footsteps, or creaking doors that play as users engage with your content.5. Community Collaboration: Share and Get Inspired!The best part about generative projects is sharing them with the community! Post your Halloween creations, get feedback, and see how others are getting into the spirit.Participate in Halloween-themed challenges or host one yourself—like a Spookiest Story Contest or Best Halloween Image Generation.Don’t forget to use the hashtag #Halloween2024 when sharing your spooky content so others can easily find and engage with your posts.6. Ethical Considerations: Keep It Fun and RespectfulWhile Halloween is all about embracing the creepy and the supernatural, it's important to remain sensitive to cultural traditions and symbols. Respectful representation goes a long way in keeping the spirit of fun alive for everyone.Ensure that your generative content is age-appropriate if targeting younger audiences—creepy doesn’t always have to mean terrifying!Happy Halloween & Happy Generating! 🎃👻We hope these tips help you create some truly terrifying (or delightfully spooky) Halloween content this year. Let your creativity run wild and embrace the eerie, the whimsical, and the downright strange. Looking forward to seeing what you conjure up this Halloween season!

1. H.R. Giger (Biomechanical Horror) Giger is famous for his nightmarish "biomechanical" art style, blending human forms with machinery and grotesque alien creatures. His designs inspired the terrifying creatures in the Alien film series, making his style a staple in sci-fi horror.2. Junji Ito (Manga Horror) Junji Ito is a Japanese manga artist known for his unsettling and disturbing imagery. His style combines detailed linework with surreal body horror, where human forms often twist, decay, or transform into unimaginable horrors.3. Francis Bacon (Abstract Horror) Bacon’s style is known for its raw and chaotic energy, often depicting distorted, screaming faces and bodies. His abstract approach creates a sense of psychological horror, focusing on human suffering and existential dread.4. Zdzisław Beksiński (Surreal Horror) Beksiński's paintings are filled with surreal, dystopian landscapes and nightmarish creatures. His style is dreamlike, featuring decaying cities, skeletal figures, and eerie, otherworldly atmospheres that evoke a sense of dread and desolation.5. Edward Gorey (Gothic Macabre) Gorey's distinctive pen-and-ink illustrations have a whimsical yet dark, gothic tone. His art features victorian-style settings, eerie characters, and morbid humor, often telling unsettling stories in a playful, minimalist way.6. Clive Barker (Fantasy Horror) Known for creating Hellraiser's Cenobites, Barker's art mixes body horror with fantasy. His style incorporates grotesque, skin-crawling depictions of demons and twisted creatures, blurring the line between pleasure and pain.7. Wayne Barlowe (Dark Fantasy) Barlowe's art focuses on the grotesque, otherworldly creatures of hellish dimensions. His works are often visually complex, mixing detailed anatomy with imaginative designs that are both disturbing and awe-inspiring.8. Dave McKean (Mixed Media Horror) McKean's style is a unique blend of photography, collage, and painting, creating eerie, surreal images that evoke fear through abstraction and texture. His works often appear in horror comics and graphic novels, including collaborations with Neil Gaiman.Each of these artists brings a distinct approach to the horror genre, using their unique styles to evoke fear, unease, or existential dread.

How to install Kohya_SS to Ubuntu WSL under Windows 111)Prepare:1.Check CPU virtualization on Windows > Task Manager > Perfomance > CPU > Virtualization: Enabled or Disabled.If Disabled - Access the UEFI (or BIOS). The way the UEFI (or BIOS) appears depends on your PC manufacturer. https://support.microsoft.com/en-us/windows/enable-virtualization-on-windows-c5578302-6e43-4b4b-a449-8ced115f58e12.Make sure you are using a recent version of Windows 10/11. If needed update to the latest version. (No earlier than Windows 10, Version 1903, Build 18362)2)Install WSL and Ubuntu1.Open Terminal > Use the command -wsl --install2.Open the Microsoft Store > Find - Ubuntu. (Ubuntu which doesn't show the version in a name is the latest)3.Install Ubuntu4.Open Ubuntu5.Create profile > For example:Username - UserPassword - User3)Install Kohya_SS on WSL Ubuntu:: Preparesudo apt updatesudo apt install software-properties-common -ysudo add-apt-repository ppa:deadsnakes/ppasudo apt updatesudo apt install python3.10 python3.10-venv python3.10-dev -ysudo apt update -y && sudo apt install -y python3-tksudo apt install python3.10-tksudo apt install git -y:: NVIDIA CUDA Toolkitwget https://developer.download.nvidia.com/compute/cuda/repos/wsl-ubuntu/x86_64/cuda-keyring_1.0-1_all.debsudo dpkg -i cuda-keyring_1.0-1_all.debsudo apt-get updatesudo apt-get -y install cudaexport PATH=/usr/local/cuda-12.6/bin${PATH:+:${PATH}}export LD_LIBRARY_PATH=/usr/local/cuda-12.6/lib64\${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}:: Rebootsudo reboot:: Kohya_ss installcd ~git clone --recursive https://github.com/bmaltais/kohya_ss.gitcd kohya_ssgit checkout sd3-sd3.5-flux./setup.sh:: Configuration settingssource venv/bin/activateaccelerate config>This machine>No distributed training

>No>No>No>All>YesIf you have a RTX 30/40 series video card choose >bf16. If don't have choose >fp16.4)Run Kohya_SS on WSL Ubuntucd kohya_ss./gui.shNotes:To find Kohya_ss folder use \\wsl.localhost\Ubuntu\home\user in Explorer. You can move there a model to train and dataset.Additional commands for Windows Terminal:Shutdown -wsl --shutdownUninstall or reset Ubuntu -wsl --unregister Ubuntu

source: Yahoo FinanceTensor.Art Becomes World's Largest VisionAI Resource Hosting Platform in Under a Year, Empowering Enterprise AIFounded in July 2023, Tensor.Art has seen its global traffic surpass 15 million in less than a year.Currently hosting over 330,000 resources and generating more than 2 million images daily, it has positioned itself as a leading generative AI service platform worldwide. Remarkably, Tensor.Art has already started to turn a profit.As a pioneering explorer of a sustainable Gen AI ecosystem, Tensor.Art provides cloud computing power for model creators and users while offering AI solutions tailored to real-world applications across various industries.Founder Shen, possessing a keen sense of computer and AI technology, quickly decided to enter the AIGC (AI-generated content) market during its early rise. This swift decision led to the establishment of a platform that offers robust support.Tensor.Art is the world's first model platform that supports online inference and online operation of full-scale models. It consistently maintains a keen insight into the latest AI technologies and promptly embraces various cutting-edge advancements, such as the globally popular Stable Diffusion3, HunYuan DiT, Kolors, Flux, and more!As one of the first to deploy StableDiffusion on the cloud, Tensor.Art maintains a keen insight into new AI technologies and rapidly integrates the latest advancements. This includes globally impactful technologies like Stable Diffusion 3, HunYuan DiT, Kolors, Flux, and more.404’s report on Tensor.Art as the world’s first company to hold AI events in 2023Operations head Sawoo states, “We are committed to providing the best platform and community for AI enthusiasts and model creators.As early as 2023, we were pioneers in the AIGC platform space, hosting diverse events and launching creator incentive programs, which have since been emulated by competitors like CivitAI.Moreover, we tirelessly promote new global technologies, ensuring rapid online integration and training capabilities.With a comprehensive community ecosystem and rich activities, Tensor.Art now leads the global growth in new foundational models, growing at 5-6 times the rate of other leading competitors, earning praise from AI enthusiasts and model creators worldwide.”A successful collaboration between Tensor.Art and SnapChatIn an effort to democratize AI and make generative services more accessible, Tensor.Art has explored numerous real-world applications.For instance, in February 2024, the platform used its AI generative capabilities in collaboration with SnapChat to create a new paradigm in creativity through AR.Subsequent partnerships include renowned tattoo artists from Austria, a famous sticker website in the UK, and an architectural firm in Turkey, offering AI-generated design inspiration.API Service：https://tams.tensor.art/Additionally, Tensor.Art is committed to serving the B2B sector by providing a GPU API platform and simplified AI tool workflows, significantly lowering the AI adoption barrier for enterprises and catering to customized needs. This makes AI services more accessible and efficient, enhancing corporate productivity and creative inspiration.Looking ahead, Tensor.Art will maintain its competitive edge by continuing to explore and quickly integrate new global technologies while also launching its own large models. This vision aims to offer an even better community experience and technical capabilities for AI enthusiasts and model creators.

Welcome to Dream Diffusion FLUX ULTIMATE, TXT 2 VID With its own custom workflow made for Tensor Arts Comfy Workspace. The workflow can be downloaded on this page....... ENJOYThis is a 2nd stage Trained checkpoint to its predecessor FLUX HYPER.When you think you had it nailed in the last version and notice a 10% margin that could still be trained........ Well that's what happened ..So now this version has even more font styles, Better adherence, Sharper image clarity and a better grasp for anime, water painting and such on....This model has the same setting parameters as Flux HyperPrompt Example : Logo in neon lights, 3D, colorful, modern, glossy, neon background,with a huge explosion of fire with epic effects, the text reads  "FLUX ULTIMATE , GAME CHANGER ",Set steps at : 20Sampler : DPM++ 2M or EULER Gives best resultsScheduler : SimpleDenoise : 1.00Image Size : 576 x 1024 or 1024 x 576 You can choose any size but this model is optimized for faster rendering with those sizes.Download the links from below and save them to your comfy folders...Comfy Workflow :  https://openart.ai/workflows/maitruclam/comfyui-workflow-for-flux-simple/iuRdGnfzmTbOOzONIiVVVae download this to your Vae folder inside of your model folderDownload them from: https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main/vaeClip:  download clip_l.safetensors and t5xxl_fp8_e4m3fn.safetensors download these 2 and save them to your clip folder inside of your models folderDownload them from : https://huggingface.co/comfyanonymous/flux_text_encoders/tree/mainIf you have any questions or issues feel free to drop a comment below and I will get back to you as soon as I can. Enjoy  DICE

I made an AI Tool yesterday ( FLUX IMG2IMG + LORAS + UPSCALE + CHOICE | ComfyUI Workflow | Tensor.Art ) that allows you to combine up to 3 LoRA's and upscale - it has model switching to let you choose whether to turn on 0/1/2/3 if the available LoRA inputs - you can choose the weighting 1 by 1 and swap out the base Flux model and all the LoRA's to your own preferences. I have implemented Radio Button prompting so that the main Trigger words for the LoRA's I use most often are already behind the buttons - and you can use "Custom" to add your own prompt or triggers into the mix.For this test I used a 6k Adobe Stock licensed image of a boat on the beach, with the Model Switcher set to "2" to prevent any bleed from other LoRA's in the tool, everything is upscaled by 4x-Ultrasharp at a factor of 2 (the tool will size your longest edge to 1024 as it processes so you will end up with a 2048 pixel final image ready for facebook servers):Original Input Image: (downsized for article)So the first test was simply putting it through the AI Tool on base Flux model - no denoise - no LoRA at all:Now I have added in the LoRA "TQ - Flux Frozen" by @TracQuoc at .9 Weight, and added .25 Denoise:Next I changed the Denoise to 0.5, you can see subtle changes, a signature has appeared, the boat is starting to change in areas and writing appearing on the side of the boat:At 0.6 Denoise the boat is starting to adapt more and the beach is changing a lot:By 0.65 you can really see dramatic changes as the boat starts to develop wheels, its almost as if the AI has a plan for this one...At 0.7 - the second boat has disappeared all together, the whole boat is on a trailer, the beach is changing into grassland:Now I am stepping to 0.01 increments as all the drama happens between .7 and .8 normally with FluxSo 0.71:0.72: (the boat is definitely changing its shape now, and you can start to see snow)0.73 you can see its becoming a land based vehicle now:0.74 it feels like a towing caravan/trailer:0.75 more detail in the towing section0.76 - everything changes and suddenly we have some kind of Safari Land Cruiser0.77 now its a camper van with a pop up roof:0.78 just some more camper style detailing but nothing dramatic:0.79 There's almost no resemblance to the original scene except sky and colours:0.8 I can't see much change here:Now I will go up in increments of 0.05 again0.85 the Frozen world has taken over, although it still has the style and colour feel of the original to some extent0.9 it's all gone (it ignored inputs over .9 and changed them back to .9)I hope you have found this a useful experiment - and will save you time and coins in playing with img2img and denoise.You can check out all my AI Tools and LoRA's on my profile here: KurtC PhotoEdLet me know if you enjoyed this and I might make some more (this was my first one).

Dear Tensorians,Thanks to the implemented feature of radio buttons for AI Tools, we can use the AI tools with much more fun now. Because I'm the one who insisted to implement it and more importantly the radio button's setting import/export features, I'll give an easy tutorial about them for the beginners. 🤗😉https://tensor.art/template/765762196352358016This is an example AI tool using radio buttons. You can see the cool radio buttons on the right. Yes! The cool thing about radio button GUI is that you don't have to remember or re-type all those crazy prompt words at all any more. You can store them in those buttons and click them! Especially if you have a very wide range of different prompting styles as most users are, you cannot even remember them all. I bet you already have your own backup memo file for those special prompts lol. Yes, we have to do it for important prompts. However, more conveniently, if you make this kind of AI tool with radio button UI, you can just store them online next to you all the time. You can click on the buttons and generate various images whenever you want, even when you are driving (just kidding, don't ever do that lol). Of course you can add extra prompt together with the buttons. (Click "custom" button and you can always input more prompt!)To create the radio buttons, click edit in the ... menu.Then you move to the EDIT page of the AI tool.in the middle of the page, you see the user-configurable settings.By clicking "Add" button, you can choose your AI interface. By clicking "edit" in the prompt's text box, you can enter the radio buttons option page.From the scratch, you can choose the pre-defined groups and buttons. In addition, you can add your own new buttons! Make a button name and its content. The content is part of prompts you want to add for the button's place.After you are done with all the button settings, click "confirm" and then "publish" your AI tool. Then you'll see your cool radio buttons in the AI tool. (Note that there are certain prompt text box nodes in comfyUI unable to edit for buttons. Basic text prompt nodes and more nodes can be used for button edit. You can check it after you publish your workflow into a tool. If it doesn't support the radio buttons, use different prompt text nodes.)Whenever you update your workflow for the AI tool, all the AI tool UI is reset to none!! Yes. It was a real headache at the beginning. However, now we have a cool import/export button for the radio buttons! (Thanks God~ 👯‍♀️⛈💯🤗). BTW, when you edit the button groups, you might choose part of the 6 or 7 groups (e.g, "character settings" and "role" groups) first and add some nice buttons, then later you change your mind and want to add another group, e.g., "style" group, however, if you press the add button for that, your previous button data will be gone!! You restart from the beginning. Be very careful! (You'll understand what I mean when it happens. lol)Before updating your workflow, you can export the radio button settings as a JSON file. Then you can import it back in later anytime you want. More importantly, you can just edit the radio buttons from the editors (like MS visual studio) for easier copy and paste from the existing files. Trust me. This will save your enormous amount of time remaking those terrible buttons all the time whenever the workflow is modified.Sometimes you must want to edit an existing button JSON file for another AI tool. Editing a JSON file is not really an entertaining work. However, it's much better than remaking the whole radio buttons at GUI~ So find the place to edit in the JSON file and change it very carefully. The JSON syntax is not very editor-friendly and error-prone. But you'll get used to it soon by trial and errors. It's always useful to use "find" command to look for the button you want in the file. You'll realize more interesting things while using the button JSON files. I'll leave them for your own pleasant surprise~ LOL.I shared my JSON file of the AI tool in the comfy-chatroom of Discord. Feel free to use it.I hope this article helped you make the radio button UI more easily. Enjoy~ 🤗😉⛈

ReviewHello everyone; I want to share some of my impressions about the Chinese model, Hunyuan-DiT from tencent. First of all let’s start with some mandatory data to know so we (westerns) can figure out what is meant for:Hunyuan-DiT works well as multi-modal dialogue with users (mainly Chinese and English language), the better explained your prompt the better your generation will be, is not necessary to introduce only keywords, despite it understands them quite well. In terms of rating HYDiT 1.2 is located between SDXL and SD3; is not as powerful than SD3, defeats SDXL almost in everything; for me is how SDXL should’ve be in first place; one of the best parts is that Hunyuan-DiT is compatible with almost all SDXL node suit.Hunyuan-DiT-v1.2, was trained with 1.5B parameters.mT5, was trained with 1.6B parameters.Recommeded VAE: sdxl-vae-fp16-fixRecommended Sampler: ddpm, ddim, or dpmmsPrompt as you’d like to do in SD1.5, don’t be shy and go further in term of length; HunyuanDiT combines two text encoders, a bilingual CLIP and a multilingual T5 encoder to improve language understanding and increase the context length; they divide your prompt on meaningful IDs and then process your entire prompt, their limit is 100 IDs or to 256 tokens. T5 works well on a variety of tasks out-of-the-box by prepending a different prefix to the input corresponding to each task.To improve your prompt, place your resumed prompt in the CLIP:TextEncoder node box (if you disabled t5), or place your extended prompt in the T5:TextEncoder node box (if you enabled t5).You can use the "simple" text encode node to only use one prompt, or you can use the regular one to pass different text to CLIP/T5.The worst is the model only benefits from moderated (high for TensorArt) step values: 40 steps are the basis in most cases.Comfyui (Comfyflow) (Example)TensorArt added all the elements to build a good flow for us; you should try it too.AdditionalWhat can we do in the Open-Source plan? (link)Official info for LoRA training (link)ReferencesAnalysis of HunYuan-DiT | https://arxiv.org/html/2405.08748v1Learn more of T5 | https://huggingface.co/docs/transformers/en/model_doc/t5How CLIP and T5 work together | https://arxiv.org/pdf/2205.11487

In the world of digital artistry, achieving that perfect blend of intricate details and stunning visuals can be a game-changer. That's where our latest model, TQ-HunYuan-More-Beautiful-Detail v1.7, comes into play. Designed with precision and a keen eye for aesthetics, this model is your go-to solution for elevating your artwork to new heights.What is TQ-HunYuan-More-Beautiful-Detail v1.7?TQ-HunYuan-More-Beautiful-Detail v1.7 is a state-of-the-art LoRA (Low-Rank Adaptation) model created to enhance the finer details in your digital creations. Whether you're working on portraits, landscapes, or abstract designs, this model ensures that every nuance and subtlety is brought to life with extraordinary clarity and beauty.Why Choose TQ-HunYuan-More-Beautiful-Detail v1.7?Unmatched Detail Enhancement: As the name suggests, this model excels at adding more beautiful details to your artwork. It meticulously enhances textures, refines edges, and highlights intricate patterns, making your creations visually striking.Versatility Across Genres: No matter the style or genre of your artwork, TQ-HunYuan-More-Beautiful-Detail v1.7 adapts seamlessly. From hyper-realistic portraits to fantastical landscapes, this model enhances every element with precision.User-Friendly Integration: Designed for ease of use, integrating TQ-HunYuan-More-Beautiful-Detail v1.7 into your workflow is straightforward. Compatible with various platforms and software, it allows artists of all levels to harness its power without a steep learning curve.Boost Your Creativity: By handling the intricate details, this model frees up your creative energy. Focus on the broader aspects of your work while TQ-HunYuan-More-Beautiful-Detail v1.7 takes care of the fine-tuning, resulting in a harmonious and polished final piece.How to Get StartedGetting started with TQ-HunYuan-More-Beautiful-Detail v1.7 is simple. Visit this link to access the model. Download and integrate it into your preferred digital art software, and watch as your creations transform with enhanced details and breathtaking beauty.Ready to take your art to the next level? Download TQ-HunYuan-More-Beautiful-Detail v1.7 now and start creating masterpieces with more beautiful detail than ever before.

SD3 understands prompts better compared to SDXL. You can use this to create interesting 3D lettering. For this purpose, use this WF! You can use a gradient as the background or any image you like. Have fun!Link to workflow: SD3 - 3D lettering designer | ComfyUI Workflow | Tensor.Art

Realistic VisionI am excited to present my latest Realistic checkpoint model based on SD3M. This model has undergone over 100k+ training steps, ensuring high-quality output.About This Model:This is a Photo Realistic model, capable of generating photorealistic images. No trigger words are needed. The model is designed to produce high-detail, high-resolution images that closely mimic real-life photographs.Configuration Used for Training:GPU: A6000x2Dataset: A mix of 5k stock photos and my own datasetBatch Size: 8Optimizer: AdamWScheduler: Cosine with restartsLearning Rate (LR): 1e-05Epoch: Target of 300 epochsCaptioning: WD14 and BLIP mixQuick Guide and Parameters:Clip Encoder: Not requiredVAE: Not requiredSampler: dpmpp_2mScheduler: sgm_uniformSampling Steps: 25+CFG Scale: 3+For better results, try using ComfyUI. Here is a workflow that is low-cost and efficient. Currently, upscaling is not possible due to specific reasons. I have reported the issue to the TA team, and hopefully, it will be fixed soon.Realistic VisionAspect Ratios for Demo:1:1 [1024x1024 square]8:5 [1216x768 landscape]4:3 [1152x896 landscape]3:2 [1216x832 landscape]7:5 [1176x840 landscape]16:9 [1344x768 landscape]21:9 [1536x640 landscape]19:9 [1472x704 landscape]3:4 [896x1152 portrait]2:3 [832x1216 portrait]5:7 [840x1176 portrait]9:16 [768x1344 portrait]9:21 [640x1536 portrait]5:8 [768x1216 portrait]9:19 [704x1472 portrait]Important: Do not include NSFW-related/mature words or censor words in your prompt. Doing so may result in unreliable or undesirable image outcomes.Note:This is not a merged or modified model. It is the original Realistic Vision fine-tuned model. Some users have been spreading incorrect information in the model's comment section. If you have any questions or want to know more, join my Discord server or share your thoughts in the comment section. Thank you for your time.

HunyuanDiT - Perfect cute animehttps://tensor.art/models/755812883138538240?source_id=nz-ypFjjk0C7pPcibn708xQiEnhance character appearance details, eyes, hair, colors, and drawings in anime styleHunyuanDiT - Realistic detailshttps://tensor.art/models/755789054659947864/HunyuanDiT-Realistic-details-V1Add more realistic details for imagesHunyuanDIT - Vivid colorhttps://tensor.art/models/755810413532312715?source_id=nz-ypFjjk0C7pPcibn708xQiEnhance vivid colors and details in photosHunyuan - Beauty Portraithttps://tensor.art/models/755789995257798458?source_id=nz-ypFjjk0C7pPcibn708xQiortrait within more details hair, skin...

EnglishToday, Ｉwill teach you how to use TensorArt to train an Hunyuan model online.Step 1: Open “Online Training.On the left side, you will see the dataset window, which is empty by default. You can upload some images to create a dataset or upload a dataset zip file. The zip file can include annotation files, following the same format as kohya-ss, where each image file corresponds to a text annotation file with the same name.In the model theme section on the right, you can choose from options such as anime characters, real people, 2.5D, standard, and custom.Here, we select “Base” and choose the Hunyuan model as the base model.For the base model parameter settings, we recommend setting the number of repetitions per image to 4 and the number of epochs to 16.、After uploading a processed dataset, if your dataset annotations include character names, you don’t need to specify a trigger word. Otherwise, you should assign a simple trigger word to your model, such as a character name or style name.Next, select an annotation file from the dataset to use as a preview prompt.If you want to use Professional Mode, click the button in the top right corner to switch to Professional Mode.In Professional Mode, it is recommended to double the learning rateand use the cosine_with_restarts learning rate scheduler. For the optimizer, you can choose AdamW8bit.Enable label shuffling and ensure that the first token remains unchanged (especially if you have a character name trigger word as the first token).Disable the noise offset feature, and you can set the convolution DIM to 8 and Alpha to 1.In the sample settings, add the Negative prompts, and then you can start the training process.In the training queue, you can view the current loss value chart and the four sample images generated for each epoch.Finally, you can choose the epoch with the best results to download to your local machine or publish directly on TensorArt.After a few minutes, your model will be deployed and ready.日本語今日、私はTensorArtを使用してHunyuanモデルをオンラインでトレーニングする方法を教えます。ステップ1: 「オンライントレーニング」を開きます。左側にデータセットウィンドウが表示され、デフォルトでは空です。データセットを作成するために画像をアップロードするか、データセットのzipファイルをアップロードできます。zipファイルには、kohya-ssと同じ形式のアノテーションファイルを含めることができ、各画像ファイルには同じ名前のテキストアノテーションファイルが対応しています。右側のモデルテーマセクションでは、アニメキャラクター、実在の人物、2.5D、標準、カスタムなどのオプションから選択できます。ここでは「Base」を選択し、Hunyuanモデルをベースモデルとして選びます。ベースモデルのパラメーター設定では、画像ごとの繰り返し回数を4、エポック数を16に設定することをお勧めします。 処理済みのデータセットをアップロードした後、データセットのアノテーションにキャラクター名が含まれている場合は、トリガーワードを指定する必要はありません。それ以外の場合は、キャラクター名やスタイル名など、モデルに簡単なトリガーワードを割り当ててください。 次に、プレビュー用プロンプトとして使用するために、データセットからアノテーションファイルを選択します。プロフェッショナルモードを使用したい場合は、右上隅のボタンをクリックしてプロフェッショナルモードに切り替えます。プロフェッショナルモードでは、学習率を倍増することをお勧めします。また、cosine_with_restarts学習率スケジューラーを使用してください。オプティマイザーとしては、AdamW8bitを選択できます。ラベルシャッフルを有効にし、最初のトークンが変更されないようにします（特にキャラクター名トリガーワードが最初のトークンの場合）。ノイズオフセット機能を無効にし、畳み込みDIMを8、Alphaを1に設定できます。サンプル設定でNegative promptsを追加し、その後、トレーニングプロセスを開始できます。トレーニングキューでは、現在の損失値チャートと各エポックごとに生成された4つのサンプル画像を表示できます。最後に、最良の結果が得られたエポックを選択して、ローカルマシンにダウンロードするか、直接TensorArtで公開できます。数分後には、モデルがデプロイされ、使用可能になります。한국인오늘은 TensorArt를 사용하여 Hunyuan 모델을 온라인에서 훈련하는 방법을 알려드리겠습니다.1단계: “온라인 훈련”을 엽니다.왼쪽에는 기본적으로 비어 있는 데이터셋 창이 표시됩니다. 데이터셋을 만들기 위해 이미지를 업로드하거나 데이터셋 zip 파일을 업로드할 수 있습니다. zip 파일에는 kohya-ss와 같은 형식의 주석 파일이 포함될 수 있으며, 각 이미지 파일에는 동일한 이름의 텍스트 주석 파일이 대응됩니다.오른쪽의 모델 테마 섹션에서는 애니메이션 캐릭터, 실제 인물, 2.5D, 표준, 사용자 정의 등 다양한 옵션 중에서 선택할 수 있습니다.여기에서는 “Base”를 선택하고 Hunyuan 모델을 기본 모델로 선택합니다.기본 모델 파라미터 설정에서는 이미지당 반복 횟수를 4로, 에포크 수를 16으로 설정하는 것을 권장합니다. 처리된 데이터셋을 업로드한 후, 데이터셋의 주석에 캐릭터 이름이 포함되어 있으면 트리거 단어를 지정할 필요가 없습니다. 그렇지 않으면 모델에 간단한 트리거 단어를 지정해야 합니다, 예를 들어 캐릭터 이름이나 스타일 이름 등. 다음으로, 미리 보기 프롬프트로 사용할 주석 파일을 데이터셋에서 선택합니다.전문 모드를 사용하려면, 오른쪽 상단의 버튼을 클릭하여 전문 모드로 전환합니다.전문 모드에서는 학습률을 두 배로 늘리는 것이 좋습니다.또한 cosine_with_restarts 학습률 스케줄러를 사용합니다. 옵티마이저로는 AdamW8bit을 선택할 수 있습니다.레이블 셔플을 활성화하고 첫 번째 토큰이 변경되지 않도록 합니다(특히 캐릭터 이름 트리거 단어가 첫 번째 토큰인 경우).노이즈 오프셋 기능을 비활성화하고, 컨볼루션 DIM을 8로, Alpha를 1로 설정할 수 있습니다.샘플 설정에서 Negative prompts를 추가한 후, 훈련 프로세스를 시작할 수 있습니다.훈련 대기열에서 현재 손실 값 차트와 각 에포크에 대해 생성된 4개의 샘플 이미지를 볼 수 있습니다.마지막으로, 가장 좋은 결과를 얻은 에포크를 선택하여 로컬 컴퓨터로 다운로드하거나 직접 TensorArt에 게시할 수 있습니다.몇 분 후, 모델이 배포되고 사용 가능해집니다.Tiếng ViệtHôm nay, tôi sẽ hướng dẫn bạn cách sử dụng TensorArt để đào tạo mô hình Hunyuan trực tuyến.Bước 1: Mở “Đào tạo trực tuyến.”Ở bên trái, bạn sẽ thấy cửa sổ tập dữ liệu, mặc định là trống. Bạn có thể tải lên một số hình ảnh để tạo tập dữ liệu hoặc tải lên tệp zip của tập dữ liệu. Tệp zip có thể bao gồm các tệp chú thích, theo cùng một định dạng như kohya-ss, trong đó mỗi tệp hình ảnh tương ứng với một tệp chú thích văn bản cùng tên.Ở phần chủ đề mô hình bên phải, bạn có thể chọn từ các tùy chọn như nhân vật anime, người thật, 2.5D, tiêu chuẩn và tùy chỉnh.Tại đây, chúng ta chọn “Base” và chọn mô hình Hunyuan làm mô hình cơ bản.Đối với cài đặt tham số của mô hình cơ bản, chúng tôi khuyên bạn nên đặt số lần lặp lại trên mỗi hình ảnh là 4 và số epoch là 16. Sau khi tải lên tập dữ liệu đã xử lý, nếu các chú thích của tập dữ liệu của bạn bao gồm tên nhân vật, bạn không cần phải chỉ định từ kích hoạt. Ngược lại, bạn nên gán một từ kích hoạt đơn giản cho mô hình của mình, chẳng hạn như tên nhân vật hoặc tên phong cách. Tiếp theo, chọn một tệp chú thích từ tập dữ liệu để sử dụng làm lời nhắc xem trước.Nếu bạn muốn sử dụng Chế độ Chuyên nghiệp, hãy nhấp vào nút ở góc trên bên phải để chuyển sang Chế độ Chuyên nghiệp.Trong Chế độ Chuyên nghiệp, nên gấp đôi tỷ lệ học.Và sử dụng bộ lập lịch tỷ lệ học cosine_with_restarts. Đối với bộ tối ưu hóa, bạn có thể chọn AdamW8bit.Kích hoạt xáo trộn nhãn và đảm bảo rằng mã thông báo đầu tiên không thay đổi (đặc biệt nếu bạn có từ kích hoạt tên nhân vật là mã thông báo đầu tiên).Tắt tính năng dịch chuyển tiếng ồn và bạn có thể đặt DIM tích chập là 8 và Alpha là 1.Trong cài đặt mẫu, thêm các Lời nhắc tiêu cực, sau đó bạn có thể bắt đầu quá trình đào tạo.Trong hàng đợi đào tạo, bạn có thể xem biểu đồ giá trị tổn thất hiện tại và bốn hình ảnh mẫu được tạo ra cho mỗi epoch.Cuối cùng, bạn có thể chọn epoch có kết quả tốt nhất để tải xuống máy tính của bạn hoặc xuất bản trực tiếp trên TensorArt.Sau vài phút, mô hình của bạn sẽ được triển khai và sẵn sàng sử dụng.españolHoy, te enseñaré cómo usar TensorArt para entrenar un modelo Hunyuan en línea.Paso 1: Abre “Entrenamiento en línea.”A la izquierda, verás la ventana del conjunto de datos, que está vacía por defecto. Puedes subir algunas imágenes para crear un conjunto de datos o subir un archivo zip del conjunto de datos. El archivo zip puede incluir archivos de anotación, siguiendo el mismo formato que kohya-ss, donde cada archivo de imagen corresponde a un archivo de anotación de texto con el mismo nombre.En la sección de temas del modelo a la derecha, puedes elegir entre opciones como personajes de anime, personas reales, 2.5D, estándar y personalizado.Aquí, seleccionamos “Base” y elegimos el modelo Hunyuan como el modelo base.Para la configuración de parámetros del modelo base, te recomendamos configurar el número de repeticiones por imagen a 4 y el número de épocas a 16. Después de subir un conjunto de datos procesado, si las anotaciones de tu conjunto de datos incluyen nombres de personajes, no necesitas especificar una palabra de activación. De lo contrario, deberías asignar una palabra de activación simple a tu modelo, como un nombre de personaje o un nombre de estilo. A continuación, selecciona un archivo de anotación del conjunto de datos para usarlo como un aviso de vista previa.Si deseas usar el Modo Profesional, haz clic en el botón en la esquina superior derecha para cambiar al Modo Profesional.En el Modo Profesional, se recomienda duplicar la tasa de aprendizaje.Y usar el programador de tasa de aprendizaje cosine_with_restarts. Para el optimizador, puedes elegir AdamW8bit.Habilita el barajado de etiquetas y asegúrate de que el primer token permanezca sin cambios (especialmente si tienes una palabra de activación de nombre de personaje como el primer token).Desactiva la función de desplazamiento de ruido y puedes configurar el DIM de convolución a 8 y Alpha a 1.En la configuración de muestra, añade los Avisos Negativos, y luego puedes comenzar el proceso de entrenamiento.En la cola de entrenamiento, puedes ver el gráfico del valor de pérdida actual y las cuatro imágenes de muestra generadas para cada época.Finalmente, puedes elegir la época con los mejores resultados para descargarla a tu máquina local o publicarla directamente en TensorArt.Después de unos minutos, tu modelo estará desplegado y listo para usar.

EnglishToday, Ｉwill teach you how to use TensorArt to train an SD3 model online.Step 1: Open “Online Training.On the left side, you will see the dataset window, which is empty by default. You can upload some images to create a dataset or upload a dataset zip file. The zip file can include annotation files, following the same format as kohya-ss, where each image file corresponds to a text annotation file with the same name.In the model theme section on the right, you can choose from options such as anime characters, real people, 2.5D, standard, and custom.Here, we select “Base” and choose the SD3 model as the base model.For the base model parameter settings, we recommend setting the number of repetitions per image to 4 and the number of epochs to 16.、After uploading a processed dataset, if your dataset annotations include character names, you don’t need to specify a trigger word. Otherwise, you should assign a simple trigger word to your model, such as a character name or style name.Next, select an annotation file from the dataset to use as a preview prompt.If you want to use Professional Mode, click the button in the top right corner to switch to Professional Mode.In Professional Mode, it is recommended to double the learning rateand use the cosine_with_restarts learning rate scheduler. For the optimizer, you can choose AdamW8bit.Enable label shuffling and ensure that the first token remains unchanged (especially if you have a character name trigger word as the first token).Disable the noise offset feature, and you can set the convolution DIM to 8 and Alpha to 1.In the sample settings, add the Negative prompts, and then you can start the training process.In the training queue, you can view the current loss value chart and the four sample images generated for each epoch.Finally, you can choose the epoch with the best results to download to your local machine or publish directly on TensorArt.After a few minutes, your model will be deployed and ready.日本語今日は、TensorArtを使用してオンラインでSD3モデルをトレーニングする方法を教えます。ステップ1: 「オンライントレーニング」を開きます。左側にデータセットウィンドウが表示され、デフォルトでは空です。データセットを作成するために画像をアップロードするか、データセットのzipファイルをアップロードできます。zipファイルには、kohya-ssと同じ形式のアノテーションファイルを含めることができ、各画像ファイルには同じ名前のテキストアノテーションファイルが対応しています。右側のモデルテーマセクションでは、アニメキャラクター、実在の人物、2.5D、標準、カスタムなどのオプションから選択できます。ここでは、「ベース」を選択し、SD3モデルをベースモデルとして選びます。ベースモデルのパラメーター設定では、画像ごとの繰り返し回数を4、エポック数を16に設定することをお勧めします。 処理済みのデータセットをアップロードした後、データセットのアノテーションにキャラクター名が含まれている場合は、トリガーワードを指定する必要はありません。それ以外の場合は、キャラクター名やスタイル名など、モデルに簡単なトリガーワードを割り当ててください。 次に、プレビュー用プロンプトとして使用するために、データセットからアノテーションファイルを選択します。プロフェッショナルモードを使用したい場合は、右上隅のボタンをクリックしてプロフェッショナルモードに切り替えます。プロフェッショナルモードでは、学習率を倍増することをお勧めします。また、cosine_with_restarts学習率スケジューラーを使用してください。オプティマイザーとしては、AdamW8bitを選択できます。ラベルシャッフルを有効にし、最初のトークンが変更されないようにします（特にキャラクター名トリガーワードが最初のトークンの場合）。ノイズオフセット機能を無効にし、畳み込みDIMを8、Alphaを1に設定できます。サンプル設定でNegative promptsを追加し、その後、トレーニングプロセスを開始できます。トレーニングキューでは、現在の損失値チャートと各エポックごとに生成された4つのサンプル画像を表示できます。最後に、最良の結果が得られたエポックを選択して、ローカルマシンにダウンロードするか、直接TensorArtで公開できます。数分後には、モデルがデプロイされ、使用可能になります。한국인오늘은 TensorArt를 사용하여 SD3 모델을 온라인으로 훈련하는 방법을 가르쳐 드리겠습니다.1단계: “온라인 훈련”을 엽니다.왼쪽에는 기본적으로 비어 있는 데이터셋 창이 표시됩니다. 데이터셋을 만들기 위해 이미지를 업로드하거나 데이터셋 zip 파일을 업로드할 수 있습니다. zip 파일에는 kohya-ss와 같은 형식의 주석 파일이 포함될 수 있으며, 각 이미지 파일에는 동일한 이름의 텍스트 주석 파일이 대응됩니다.오른쪽의 모델 테마 섹션에서는 애니메이션 캐릭터, 실제 인물, 2.5D, 표준, 사용자 정의 등 다양한 옵션 중에서 선택할 수 있습니다.여기에서는 “Base”를 선택하고 SD3 모델을 기본 모델로 선택합니다.기본 모델 파라미터 설정에서는 이미지당 반복 횟수를 4로, 에포크 수를 16으로 설정하는 것을 권장합니다. 처리된 데이터셋을 업로드한 후, 데이터셋의 주석에 캐릭터 이름이 포함되어 있으면 트리거 단어를 지정할 필요가 없습니다. 그렇지 않으면 모델에 간단한 트리거 단어를 지정해야 합니다, 예를 들어 캐릭터 이름이나 스타일 이름 등. 다음으로, 미리 보기 프롬프트로 사용할 주석 파일을 데이터셋에서 선택합니다.전문 모드를 사용하려면, 오른쪽 상단의 버튼을 클릭하여 전문 모드로 전환합니다.전문 모드에서는 학습률을 두 배로 늘리는 것이 좋습니다.또한 cosine_with_restarts 학습률 스케줄러를 사용합니다. 옵티마이저로는 AdamW8bit을 선택할 수 있습니다.레이블 셔플을 활성화하고 첫 번째 토큰이 변경되지 않도록 합니다(특히 캐릭터 이름 트리거 단어가 첫 번째 토큰인 경우).노이즈 오프셋 기능을 비활성화하고, 컨볼루션 DIM을 8로, Alpha를 1로 설정할 수 있습니다.샘플 설정에서 Negative prompts를 추가한 후, 훈련 프로세스를 시작할 수 있습니다.훈련 대기열에서 현재 손실 값 차트와 각 에포크에 대해 생성된 4개의 샘플 이미지를 볼 수 있습니다.마지막으로, 가장 좋은 결과를 얻은 에포크를 선택하여 로컬 컴퓨터로 다운로드하거나 직접 TensorArt에 게시할 수 있습니다.몇 분 후, 모델이 배포되고 사용 가능해집니다.Tiếng ViệtHôm nay, tôi sẽ hướng dẫn bạn cách sử dụng TensorArt để huấn luyện mô hình SD3 trực tuyến.Bước 1: Mở “Đào tạo trực tuyến.”Ở bên trái, bạn sẽ thấy cửa sổ tập dữ liệu, mặc định là trống. Bạn có thể tải lên một số hình ảnh để tạo tập dữ liệu hoặc tải lên tệp zip của tập dữ liệu. Tệp zip có thể bao gồm các tệp chú thích, theo cùng một định dạng như kohya-ss, trong đó mỗi tệp hình ảnh tương ứng với một tệp chú thích văn bản cùng tên.Ở phần chủ đề mô hình bên phải, bạn có thể chọn từ các tùy chọn như nhân vật anime, người thật, 2.5D, tiêu chuẩn và tùy chỉnh.Tại đây, chúng ta chọn “Cơ bản” và chọn mô hình SD3 làm mô hình cơ sở.Đối với cài đặt tham số của mô hình cơ bản, chúng tôi khuyên bạn nên đặt số lần lặp lại trên mỗi hình ảnh là 4 và số epoch là 16. Sau khi tải lên tập dữ liệu đã xử lý, nếu các chú thích của tập dữ liệu của bạn bao gồm tên nhân vật, bạn không cần phải chỉ định từ kích hoạt. Ngược lại, bạn nên gán một từ kích hoạt đơn giản cho mô hình của mình, chẳng hạn như tên nhân vật hoặc tên phong cách. Tiếp theo, chọn một tệp chú thích từ tập dữ liệu để sử dụng làm lời nhắc xem trước.Nếu bạn muốn sử dụng Chế độ Chuyên nghiệp, hãy nhấp vào nút ở góc trên bên phải để chuyển sang Chế độ Chuyên nghiệp.Trong Chế độ Chuyên nghiệp, nên gấp đôi tỷ lệ học.Và sử dụng bộ lập lịch tỷ lệ học cosine_with_restarts. Đối với bộ tối ưu hóa, bạn có thể chọn AdamW8bit.Kích hoạt xáo trộn nhãn và đảm bảo rằng mã thông báo đầu tiên không thay đổi (đặc biệt nếu bạn có từ kích hoạt tên nhân vật là mã thông báo đầu tiên).Tắt tính năng dịch chuyển tiếng ồn và bạn có thể đặt DIM tích chập là 8 và Alpha là 1.Trong cài đặt mẫu, thêm các Lời nhắc tiêu cực, sau đó bạn có thể bắt đầu quá trình đào tạo.Trong hàng đợi đào tạo, bạn có thể xem biểu đồ giá trị tổn thất hiện tại và bốn hình ảnh mẫu được tạo ra cho mỗi epoch.Cuối cùng, bạn có thể chọn epoch có kết quả tốt nhất để tải xuống máy tính của bạn hoặc xuất bản trực tiếp trên TensorArt.Sau vài phút, mô hình của bạn sẽ được triển khai và sẵn sàng sử dụng.españolHoy, les enseñaré cómo utilizar TensorArt para entrenar un modelo SD3 en línea.Paso 1: Abre “Entrenamiento en línea.”A la izquierda, verás la ventana del conjunto de datos, que está vacía por defecto. Puedes subir algunas imágenes para crear un conjunto de datos o subir un archivo zip del conjunto de datos. El archivo zip puede incluir archivos de anotación, siguiendo el mismo formato que kohya-ss, donde cada archivo de imagen corresponde a un archivo de anotación de texto con el mismo nombre.En la sección de temas del modelo a la derecha, puedes elegir entre opciones como personajes de anime, personas reales, 2.5D, estándar y personalizado.Aquí, seleccionamos “Base” y elegimos el modelo SD3 como el modelo base.Para la configuración de parámetros del modelo base, te recomendamos configurar el número de repeticiones por imagen a 4 y el número de épocas a 16. Después de subir un conjunto de datos procesado, si las anotaciones de tu conjunto de datos incluyen nombres de personajes, no necesitas especificar una palabra de activación. De lo contrario, deberías asignar una palabra de activación simple a tu modelo, como un nombre de personaje o un nombre de estilo. A continuación, selecciona un archivo de anotación del conjunto de datos para usarlo como un aviso de vista previa.Si deseas usar el Modo Profesional, haz clic en el botón en la esquina superior derecha para cambiar al Modo Profesional.En el Modo Profesional, se recomienda duplicar la tasa de aprendizaje.Y usar el programador de tasa de aprendizaje cosine_with_restarts. Para el optimizador, puedes elegir AdamW8bit.Habilita el barajado de etiquetas y asegúrate de que el primer token permanezca sin cambios (especialmente si tienes una palabra de activación de nombre de personaje como el primer token).Desactiva la función de desplazamiento de ruido y puedes configurar el DIM de convolución a 8 y Alpha a 1.En la configuración de muestra, añade los Avisos Negativos, y luego puedes comenzar el proceso de entrenamiento.En la cola de entrenamiento, puedes ver el gráfico del valor de pérdida actual y las cuatro imágenes de muestra generadas para cada época.Finalmente, puedes elegir la época con los mejores resultados para descargarla a tu máquina local o publicarla directamente en TensorArt.Después de unos minutos, tu modelo estará desplegado y listo para usar.

首先点击右上角的头像，在弹出的下拉框中选择我训练的模型，进入训练中心。如果之前有训练过模型，这里会看到许多训练任务。然后选择在线训练按钮进行一次训练。左侧是数据集窗口，默认没有任何数据。您可以上传一些图片作为数据集，或者上传一个数据集压缩包，压缩包可以包含标注文件，格式和kohya-ss一样，每个图片文件对应一个同名的标注文件txt。右边的模型主题中可以选择二次元人物、真实人物、2.5D、标准以及自定义。训练混元模型这里我们选择标准，在使用底模中选择混元1.2模型。混元模型使用了40depth的块，所以非常大，训练相对速度较慢，需要更高的学习率，默认使用4e-4，默认单张图片重复次数5，优化器AdamW。基础模式下参数选择，推荐单张图片重复次数5，轮数为16。上传一个处理好的数据集后，如果你的数据集标注中有人物名，可以不写触发词。否则你应该给你的模型起一个简单的触发词，例如人物名称或者风格名称。接着从数据集中选择一个标注文件作为预览提示词。如果你想使用专业模式，选择右上角按钮切换到专业模式。专业模式推荐学习率翻倍，然后使用cosine_with_restarts学习率调度器，优化器选择AdamW或者AdamW8bit。开启打乱标签(shuffle)，并且保持第1个token(如果你有一个人名触发词在第一个)关闭噪声偏移功能，卷积DIM和Alpha可以选择8和1。在样图设置中追加填写反向提示词，接下来就可以开始训练了。在训练队列中，你可以看到当前loss值变化表以及每轮epoch产生的4张样图。最后可以选择效果最好的epoch下载到本地或者直接在tensorart上发布。

SD3 can generate interesting images, but it has a huge problem with the human body. However, I noticed that simply reducing the image size to 60% can, in most cases, eliminate issues with image composition as well as extra hands or legs. This workflow does not solve the problem of having six fingers, etc. :)Base model: https://tensor.art/models/751330255836302856/Aderek-SD3-v1 or https://civitai.com/models/600179/aderek-sd3Look at the image below. You might say: "Hey, nothing's wrong here." Well, that's because you're already seeing the generation based on the reduced size. Below, you have the original image.Use composition on to use this trick&tips.Have fun!Support Paweł Tomczuk on Ko-fi! ❤️. ko-fi.com/aderek514 - Ko-fi ❤️ Where creators get support from fans through donations, memberships, shop sales and more! The original 'Buy Me a Coffee' Page.Visit my DeviantArt page: Aderek - Hobbyist, Digital Artist | DeviantArt

Exception (routeId: 7544339967855538950230)Suspect nodes:<string function>. <LayeStyle>, <LayerUtility>, <FaceDetailer>, many <TextBox>, <Bumpmap>After some reseach (on my own) I've found<FaceDetailer> node is completely broken<TextBox> and <MultiLine:Textbox> node will cause this error if you introduce more than 250+ characters, I'm not very sure about this number, but you won't be able to introduce a decent amount of text anymore.More than 40 nodes, despite its function will couse this error.How do i know this? Well I made a functional comfyflow following those rules:https://tensor.art/template/754955251181895419The next functional comfyflow suddelny stopped from generating, it's almost the same flow than the previous, but with <FaceDetailer> and large text strings to polish the prompt. It works again yay!https://tensor.art/template/752678510492967987 proof it really worked (here)I feel bad for you if this error suddenly disrupt your day; feel bad for me cuz I bought the yearly membership of this broken product I can't refound. I'll be happy to delete this bad review if you fix this error.News081124 | <String Function> has been taken down. Comfyflow works slowly (but works)081024 | eveything is broken again lmao, we cant generate outside TAMS.080624 | <reroute> output node could trigger this error when linked to many inputs.072824 | <FaceDetailer> node seems to work again.

Hello again! In this little article I want to explain the upscaling methods that I know in ComfyUI and that I have researched. I hope they will help you and that you can use them in the creation of your workflows and AI tools. In addition, remember that if you have any useful knowledge, you can share it in the comments section to enrich the topic. Also, please excuse any spelling mistakes; I am just learning English hehe.¡Let’s get to the point!To the best of my knowledge, there are two widely used ways in ComfyUI to achieve uspcaling (you decide which one to use according to your needs). The two options are: Algorithm Method or Latent Method.Algorithm Method:This is one of the most commonly used method, and is readily available. It consists of loading an upscaling model, and connecting it to the workflow. That way the image pixels are manipulated as the user wishes. It is very similar to the upscale method used in the normal way of creating images in Tensor Art.The following nodes are needed:A. Load Upscale Model.B. Upscale Image (Using Model).These nodes are connected to the workflow between the “VAE Decode” and “Save Image” nodes; as shown in the image. Once this structure is created, you can choose from all the different models offered by the “Load Upscale Model” node, ranging from “2x-ESRGAN.pth” to “SwimIR_4x”. You can use any of the 23 available models and experiment with any of them. You just have to click on the node and the list will be displayed.This can also be achieved in other ways by using another node such as “Upscale Image By”. The structure is simpler to create because only that node is connected between the VAE decode and Save Image as shown in the following image.Once the node is connected, you are free to select the mode in which you want to upscale the image (Upscale_method) and you can also set the scale to which you can recondition the image pixel value (Scale By).Strengths and Weaknesses of the Algorithm Method:Among the strengths of this method are its ease of integration into the workflow and its advantage of choosing between several upscaling model options. It also allows fast generation both in the ComfyUI and in the use of AI tools.However, among its weaknesses, it is not very effective in some specific contexts. For example: the algorithm can upscale the image pixels but does not alter the actual image size; causing the generated image itself to end up being blurred in some cases.Latent Method:This is the other alternative option to the algorithm method. It is focused on highlighting image details and maximizing quality. This method is also one of the most used in the Workflow mode of different visual content creation platforms with artificial intelligence. Here, upscaling is performed while the image is generated from latent space (Latent space is where the IA takes data from the prompt, deconstructs it for analysis and then reconstructs it to represent it in an image).The Latent Upscale node is placed between the two Ksamplers. While the first Ksampler is connected to the “Empty Latent Image” node, the second one is connected to the “VAE Decode” to ensure the correct processing and representation of the generated image.It should be noted that the “Empty Latent Image” node and the “VAE Decode” node are already included by default in the Text2Image templates in WorkFlow mode. (For more information about Text2Image, you can see my other article called “ComfyUi: Text2Image Basic Glossary”).It is important to take into consideration that for this method to work properly, you have to know how to create a correct balance between the original size of the image and its upscaled size. For example, you can generate a 512x512 image and upscale it to 1024x1024; but it is not recommended to make a 512x512 image (square image) and upscale it to 768x11152 (rectangular image) since the shape of the image would not be compatible with its uspcale version. For this reason you have to pay attention to the values of the “Empty Latent Image” and the “Latent Upscale”, so that these are always proportional.In the “Empty Latent Image” node you must place the original image dimensions (for example: 768x1152); while in the “Latent Upscale” node you must place the resized image dimensions (for example: 1152x1728). In this way you are given the freedom to set the image size to your own discretion. For this I always recommend to look at the size and upscale of the normal mode in which we create illustrations, this way we will always know which values to set and which will be compatible. As you can see in the image. You look at those values, and then write them to the nodes listed above.Once everything is connected and configured, you are able to have images of any size you want. You can experiment to your taste.Strengths and Weaknesses of the Latent Method:As strengths this option should be highlighted that it allows you to access excellent quality images if everything is correctly configured. It also allows you to create images of a custom size and upscale with the values you want. It brings out the details in both SD and XL images.As negative points we have to configure everything manually every time you want to change the size of the images or the shape of the same. Also, this method is just a little bit slower in the generation process compared to the algorithm method.Which is better: ¿Algorithm or Latent?Neither method is better than the other. Both are useful in different contexts. Remember that workflows will be different from user to user, because we all have different ways of creating and designing things.It all depends on your taste and whether you want something simpler or more elaborate. I hope the explanation in this article has helped you to make Workflows more complex and to make it easier to make the images you want.Extra Tip:If you do not find any of the nodes outlined in this document. You can double click on any empty place in the workflow and you can search for the name of the node you are looking for. Just remember to type the name without spaces.

Today, I noticed that I can add ControlNet to the SD3 model.The Tiled function works very well, so I incorporated it into my workflow and created a group for generating artistic images based on a given photo or a previously generated image. In the main part of the workflow, I simply set a very short prompt, like "grass, flowers," and I get an image that blends grass and flowers in an arrangement resembling the base photo.https://youtu.be/sv35wKNiFGsControlnet with SD3 | ComfyUI Workflow | Tensor.Art

首先点击右上角的头像，在弹出的下拉框中选择我训练的模型，进入训练中心。如果之前有训练过模型，这里会看到许多训练任务。然后选择在线训练按钮进行一次训练。左侧是数据集窗口，默认没有任何数据。您可以上传一些图片作为数据集，或者上传一个数据集压缩包，压缩包可以包含标注文件，格式和kohya-ss一样，每个图片文件对应一个同名的标注文件txt。右边的模型主题中可以选择二次元人物、真实人物、2.5D、标准以及自定义。这里我们选择自定义，在使用底模中选择SD3模型。注意在选择版本中下拉框内选择T5XXL的版本，这样才可以训练T5文本编码器。基础模式下参数选择，推荐单张图片重复次数4，轮数为16。上传一个处理好的数据集后，如果你的数据集标注中有人物名，可以不写触发词。否则你应该给你的模型起一个简单的触发词，例如人物名称或者风格名称。接着从数据集中选择一个标注文件作为预览提示词。如果你想使用专业模式，选择右上角按钮切换到专业模式。专业模式推荐学习率翻倍，然后使用cosine_with_restarts学习率调度器，优化器可以选择AdamW8bit。开启打乱标签(shuffle)，并且保持第1个token(如果你有一个人名触发词在第一个)关闭噪声偏移功能，卷积DIM和Alpha可以选择8和1。在样图设置中追加填写反向提示词，接下来就可以开始训练了。在训练队列中，你可以看到当前loss值变化表以及每轮epoch产生的4张样图。最后可以选择效果最好的epoch下载到本地或者直接在tensorart上发布。

So first, you need to update your version of OneTrainer.Second, u need dowload ALL files and folders (and rename)stabilityai/stable-diffusion-3-medium-diffusers at main (huggingface.co)then u put it:With float16 output lora has only 36MB:This is my setting for a style training:My checkpoint to testing u can dowload for free:Aderek SD3 - v1 | Stable Diffusion Model - Checkpoint | Tensor.Artand my loras: Aderek514's Profile | Tensor.ArtSo, good luck!

ReActor Node for ComfyUI 👉Downlond👈 https://github.com/lingkops4/lingko-FaceReActor-Nodeworkflowhttps://github.com/lingkops4/lingko-FaceReActor-Node/blob/main/face_reactor_workflows.jsonThe Fast and Simple Face Swap Extension Node for ComfyUI, based on ReActor SD-WebUI Face Swap ExtensionThis Node goes without NSFW filter (uncensored, use it on your own responsibility)| Installation | Usage | Troubleshooting | Updating | Disclaimer | Credits | Note!✨What's new in the latest update✨💡0.5.1 ALPHA1Support of GPEN 1024/2048 restoration models (available in the HF dataset https://huggingface.co/datasets/Gourieff/ReActor/tree/main/models/facerestore_models)👈[]~(￣▽￣)~*ReActorFaceBoost Node - an attempt to improve the quality of swapped faces. The idea is to restore and scale the swapped face (according to the face_size parameter of the restoration model) BEFORE pasting it to the target image (via inswapper algorithms), more information is here (PR#321)InstallationSD WebUI: AUTOMATIC1111 or SD.NextStandalone (Portable) ComfyUI for WindowsUsageYou can find ReActor Nodes inside the menu ReActor or by using a search (just type "ReActor" in the search field)List of Nodes:••• Main Nodes •••💡ReActorFaceSwap (Main Node Download)👈[]~(￣▽￣)~*ReActorFaceSwapOpt (Main Node with the additional Options input)ReActorOptions (Options for ReActorFaceSwapOpt)ReActorFaceBoost (Face Booster Node)ReActorMaskHelper (Masking Helper)••• Operations with Face Models •••ReActorSaveFaceModel (Save Face Model)ReActorLoadFaceModel (Load Face Model)ReActorBuildFaceModel (Build Blended Face Model)ReActorMakeFaceModelBatch (Make Face Model Batch)••• Additional Nodes •••ReActorRestoreFace (Face Restoration)ReActorImageDublicator (Dublicate one Image to Images List)ImageRGBA2RGB (Convert RGBA to RGB)Connect all required slots and run the query.Main Node Inputsinput_image - is an image to be processed (target image, analog of "target image" in the SD WebUI extension);Supported Nodes: "Load Image", "Load Video" or any other nodes providing images as an output;source_image - is an image with a face or faces to swap in the input_image (source image, analog of "source image" in the SD WebUI extension);Supported Nodes: "Load Image" or any other nodes providing images as an output;face_model - is the input for the "Load Face Model" Node or another ReActor node to provide a face model file (face embedding) you created earlier via the "Save Face Model" Node;Supported Nodes: "Load Face Model", "Build Blended Face Model";Main Node OutputsIMAGE - is an output with the resulted image;Supported Nodes: any nodes which have images as an input;FACE_MODEL - is an output providing a source face's model being built during the swapping process;Supported Nodes: "Save Face Model", "ReActor", "Make Face Model Batch";Face RestorationSince version 0.3.0 ReActor Node has a buil-in face restoration.Just download the models you want (see Installation instruction) and select one of them to restore the resulting face(s) during the faceswap. It will enhance face details and make your result more accurate.Face IndexesBy default ReActor detects faces in images from "large" to "small".You can change this option by adding ReActorFaceSwapOpt node with ReActorOptions.And if you need to specify faces, you can set indexes for source and input images.Index of the first detected face is 0.You can set indexes in the order you need.E.g.: 0,1,2 (for Source); 1,0,2 (for Input).This means: the second Input face (index = 1) will be swapped by the first Source face (index = 0) and so on.GendersYou can specify the gender to detect in images.ReActor will swap a face only if it meets the given condition.💡Face ModelsSince version 0.4.0 you can save face models as "safetensors" files (stored in ComfyUI\models\reactor\faces) and load them into ReActor implementing different scenarios and keeping super lightweight face models of the faces you use.To make new models appear in the list of the "Load Face Model" Node - just refresh the page of your ComfyUI web application.(I recommend you to use ComfyUI Manager - otherwise you workflow can be lost after you refresh the page if you didn't save it before that).TroubleshootingI. (For Windows users) If you still cannot build Insightface for some reasons or just don't want to install Visual Studio or VS C++ Build Tools - do the following:(ComfyUI Portable) From the root folder check the version of Python:run CMD and type python_embeded\python.exe -VDownload prebuilt Insightface package for Python 3.10 or for Python 3.11 (if in the previous step you see 3.11) or for Python 3.12 (if in the previous step you see 3.12) and put into the stable-diffusion-webui (A1111 or SD.Next) root folder (where you have "webui-user.bat" file) or into ComfyUI root folder if you use ComfyUI PortableFrom the root folder run:(SD WebUI) CMD and .\venv\Scripts\activate(ComfyUI Portable) run CMDThen update your PIP:(SD WebUI) python -m pip install -U pip(ComfyUI Portable) python_embeded\python.exe -m pip install -U pip💡Then install Insightface:(SD WebUI) pip install insightface-0.7.3-cp310-cp310-win_amd64.whl (for 3.10) or pip install insightface-0.7.3-cp311-cp311-win_amd64.whl (for 3.11) or pip install insightface-0.7.3-cp312-cp312-win_amd64.whl (for 3.12)(ComfyUI Portable) python_embeded\python.exe -m pip install insightface-0.7.3-cp310-cp310-win_amd64.whl (for 3.10) or python_embeded\python.exe -m pip install insightface-0.7.3-cp311-cp311-win_amd64.whl (for 3.11) or python_embeded\python.exe -m pip install insightface-0.7.3-cp312-cp312-win_amd64.whl (for 3.12)Enjoy!II. "AttributeError: 'NoneType' object has no attribute 'get'"This error may occur if there's smth wrong with the model file inswapper_128.onnx💡Try to download it manually from here and put it to the ComfyUI\models\insightface replacing existing oneIII. "reactor.execute() got an unexpected keyword argument 'reference_image'"This means that input points have been changed with the latest updateRemove the current ReActor Node from your workflow and add it againIV. ControlNet Aux Node IMPORT failed error when using with ReActor NodeClose ComfyUI if it runsGo to the ComfyUI root folder, open CMD there and run:python_embeded\python.exe -m pip uninstall -y opencv-python opencv-contrib-python opencv-python-headlesspython_embeded\python.exe -m pip install opencv-python==4.7.0.72That's it!reactor+controlnetV. "ModuleNotFoundError: No module named 'basicsr'" or "subprocess-exited-with-error" during future-0.18.3 installationDownload https://github.com/Gourieff/Assets/raw/main/comfyui-reactor-node/future-0.18.3-py3-none-any.whlPut it to ComfyUI root And run:python_embeded\python.exe -m pip install future-0.18.3-py3-none-any.whlThen:python_embeded\python.exe -m pip install basicsrVI. "fatal: fetch-pack: invalid index-pack output" when you try to git clone the repository"Try to clone with --depth=1 (last commit only):git clone --depth=1 https://github.com/Gourieff/comfyui-reactor-nodeThen retrieve the rest (if you need):git fetch --unshallow

Hello! This is my first article; I hope it will be of benefit to the person who reads it. I still have limited knowledge about WorkFlow; but I have researched and learned little by little. If anyone would like to contribute some content; you are totally free to do so. Thank you.I made this article to give a brief and basic explanation about basic concepts about Comfyui or WorkFlow. This is a technology with many possibilities and it would be great to make it easier to use for everyone! What is Workflow?Workflow is one of the two main image generation systems that Tensor Art has at the moment. It corresponds to a generation method that is characterized by a great capacity to stimulate the creativity of the users; also, it allows us to access to some Pro features being Free users.How do I access the WorkFlow mode?To access the WorkFlow mode, you must place the mouse cursor on the “Create” tab as if you were going to create an image by conventional means. Once you have done that; click on the “ComfyFlow” option and you are done.After that, you will see a tab with two options “New WorkFlow” and “Import WorkFlow”. The first one allows you to start a workflow from a template or from scratch; while the second option allows you to load a workflow that you have saved on your pc in a JSON file.If you click on the “New WorkFlow” option, a tab with a list of various templates will be displayed (each template will have a different purpose). But the main one will be “Text2Image”; it will allow us to create images from text, similarly to the conventional method we always use. You can also create a workflow from scratch in the “Empty WorkFlow Template” option but for a better explanation of the basics we will use the “Text2Image”.Once you click on the "Text2Image" option, you must wait a few seconds and a new tab will be displayed with the template, which contains the basics to create an image by means of text. Nodes and Borders: ¿What are they and how do they work?Well, to understand the basics of how a WorkFlow works, it is necessary to have a clear understanding of what Nodes and Border are.Nodes are small boxes that are present in the workflow; each node will have a specific function necessary for the creation, enhancement or editing of the image or video. The basics of Text2Image are the CheckPoint loader, the Clip Text Encoders, the Empty Lantent Image, the Ksampler, the VAE decoder, and Save Image. It should be noted that there are hundreds of other nodes besides these basics and they all have many different functions.On the other hand, the “Borders” are the small colored wires that connect the different nodes. They are the ones that will set which nodes will be directly related. The Borders are ordered by colors that are generally related to a specific function.The purple is related to the Model or Lora used.The yellow one is intended for connection to the model or lora with the space to place the prompt.The red refers to VAE.The orange color refers to the connection between the spaces for placing the prompt and the “Ksampler” node.The fucsia color makes allusion to the latent, which will serve for many things; but for this case it serves to connect the “Empty Latent Image” node with the “Ksampler” node and establish the number and size of the images that will be generated.And the blue color is related to everything that has to do with images; it has many uses but this case is related to the “Save Image” node.What are the Text2Image template Nodes used for?Having this clear is of utmost relevance, since it allows you to know what each node of this basic template is for. It's like knowing what each piece in a lego set is for and understanding how they should be connected to create a beautiful masterpiece! Also, if you get to know what these nodes are for, it will be easier for you to intuit the functionality of its variants and other derived nodes.A) The first one is the node called “Load Chckpoint”, this node has three specific functions. The first one is to load the base model or checkpoint with which an image will be created. The second is the Clip, which will take care of connecting the positive and negative prompts that you write to the checkpoint. And the third is that it connects and helps to load the VAE model. B) The second one is the “Empty Latent Image”; which is the node in charge of processing the image dimensions from the latent space. It has two functions: First, set the width and length of the image; and second, set how many images will be generated simultaneously according to the “Batch Size” option.C) The third is the two “Clip Text Enconder” nodes: in this case there will always be at least two of these nodes, since they are responsible for setting both the positive and negative prompts that you write to describe the image you want. They are usually connected to the "Load Checkpoint" or any LoRa and are also connected to the “Ksampler” node.D) Then, there is a node “Ksampler”. This node is the central point of all WorkFlow; it is the one that sets the most important parameters in the creation of images. It has several functions: the first one is to determine which is the seed of the image and to regulate how much it changes from image to generated image by means of the “control_after_generate” option. The second function is to set how many steps are needed to create the image (you set them as you wish); the third function is to determine which sampling method is used and also what is the scheduler of this method (this helps to regulate how much space is eliminated when creating the image).E) The penultimate one is the VAE decoder. This node is in charge of assisting the processing of the image to be generated: its main function is to be responsible for materializing the written prompt into an image. That is to say, it reconstructs the description of the image we want as one of the final steps to finish the generation process. Then, the information is transmitted to the “Save Image” node to display the generated image as the final product.F) The last node to explain is the “Save Image”. This node has the simple function of saving the generated image and providing the user with a view of the final work that will later be stored in the taskbar where all the generated images are located.Final Consideration:This has been a small summary and explanation about very basic concepts about ComfyUI Mode; you could even say that it is like a small glossary about general terms. I have tried to give a small notion that tries to facilitate the understanding of this image generation tool. There is still a lot to explain, but I will try to cover all the topics; the information would not fit in a single article (ComfyUI is a whole universe of possibilities). ¡Thank you so much for taking the time to read this article!

ComfyUI_examplesTextual Inversion Embeddings ExamplesHere is an example for how to use Textual Inversion/Embeddings.To use an embedding put the file in the models/embeddings folder then use it in your prompt like I used the SDA768.pt embedding in the previous picture.Note that you can omit the filename extension so these two are equivalent:embedding:SDA768.ptembedding:SDA768You can also set the strength of the embedding just like regular words in the prompt:(embedding:SDA768:1.2)Embeddings are basically custom words so where you put them in the text prompt matters.For example if you had an embedding of a cat:red embedding:catThis would likely give you a red cat.

Art MediumsVarious art mediums. Prompted with '{medium} art of a woman MetalpointMiniature PaintingMixed MediaMonotype PrintingMosaic Tile ArtMosaicNeonOil PaintOrigamiPapermakingPapier-mâchéPastelPen And InkPerformance ArtPhotographyPhotomontagePlasterPlastic ArtsPolymer ClayPrintmakingPuppetryPyrographyQuillingQuilt ArtRecycled ArtRelief PrintingResinReverse Glass PaintingSandScratchboard ArtScreen PrintingScrimshawSculpture WeldingSequin ArtSilk PaintingSilverpointSound ArtSpray PaintStained GlassStencilStoneTapestryTattoo ArtTemperaTerra-cottaTextile ArtVideo ArtVirtual Reality ArtWatercolorWaxWeavingWire SculptureWoodWoodcutGlassGlitch ArtGold LeafGouacheGraffitiGraphite PencilIceInk Wash PaintingInstallation ArtIntaglio PrintingInteractive MediaKinetic ArtKnittingLand ArtLeatherLenticular PrintingLight ProjectionLithographyMacrameMarbleMetalColored PencilComputer-generated Imagery (cgi)Conceptual ArtCopper EtchingCrochetDecoupageDigital MosaicDigital PaintingDigital SculptureDioramaEmbroideryEnamelEncaustic PaintingEnvironmental ArtEtchingFabricFeltingFiberFoam CarvingFound ObjectsFrescoAugmented Reality ArtBatikBeadworkBody PaintingBookbindingBronzeCalligraphyCast PaperCeramicsChalkCharcoalClayCollageCollagraphy3d PrintingAcrylic PaintAirbrushAlgorithmic ArtAnimationArt GlassAssemblage

SD3 Anime LoRA is Finally Here!I am thrilled to announce that the SD3 Anime LoRA model is finally available. In addition, I am releasing a new update that includes an SD3 anime checkpoint model.Currently, I am publishing a beta version as I continue to work diligently to perfect the model. I aim to have the final release ready by the end of this month or early August.Stay tuned, as the SD3 Anime beta version will be available within the next couple of days!Here are some guidelines to use this LoRA to its full potential:If you are trying to create any specific subject or object, use trigger word like 'anime style' in your prompt.If you're targeting a character, you can ignore the keyword and go with something like this:For a male character: 'anime boy'For a female character: 'anime girl'Simple, right? You can also use the trigger word 'anime style' most of the time. I've noticed it gives better results.ModelRecommended Parameter :LoRA Weight : 0🆙1VAE : No NeedSampler : DPM++ 2M SGM UniformSteps : 20➡30CFG : 3➡4Upscaler : R-ESRGAN 4x+If you encounter any issues, I recommend using ComfyUI for a better experience. Here's the workflow: ComfyUI Workflow. Open the link, select the LoRA model, choose the LoRA strength, and hit the run button.Join my community, Share your feedback, learn, and have fun with us! 😊Discord➡️https://discord.gg/QQKd7bu97P

Hello everyone! ✨ Today I will bring you a super practical tutorial: How to set up a convenient prompt word radio version for your AI Tools! 😎 Save it quickly, and you will never have to worry about how to set prompt words again! 👌Are you ready for the course? Let's get started! 🔍First, the first step is to open the official website of TensorArt. 📂 After opening, you will see a variety of AI tools and resources, which are very rich~ 👀Next, open comfyflow and start making our AI Tool! 🤖 This process is simple and fun, let's explore it together! ✨In comfyflow, we click the "New" button, which will take you to a new interface~ 🖱️💻In this interface, we can start creating our own workflow~ 🌟🎉 Next, we need to fill in the positive prompt words, which is a super critical step! 📝✨In the positive prompt word area, we need to enter the content we want. 📋 Here, the editor simply wrote an example for everyone: "a man". 🤵 This example is just for the convenience of teaching, you can freely play according to your needs~ 🌈🎆🎉 When you have completed the workflow, you can click the "Publish" button in the upper right corner! 🚀✨Don't forget to give your AI Tool an interesting name! 💡 This name will make your tool more attractive~✨ In addition, remember to divide the area correctly, so that you can see it clearly and it is also convenient for your friends to find and use it! 📂🔍🌟 Next, let's complete the next step together! 💪We pull down the current interface and find the user-configurable settings area. 👏 Then click the "Add" button. This step is very critical! 🖱️✨ Everyone must remember to add your positive prompt word node! 🔍✨After adding the node, our next step is to click the "Set" button on the right to proceed to the next step. 🔧✨ This step is crucial! Don't miss it! 😉🚀✨ The next step is also very important! 😊First, click the radio button, then click "Add". 🔘✨ Here, you can add the buttons you want to release to the user! 👍 After selecting, be sure to click "Confirm"! ✔️✨Friends, we have finally reached the last step! 🎉💪 This is an exciting moment! ✨When you have completed all the operations, remember to click the "Publish" button to publish your AI gadget! 🚀✨ Can't wait to see the results? Hurry up and generate a picture yourself to try and experience your results! 🌟🖼️Well, that's all for today's tutorial! 😊 I hope everyone can complete it successfully and create their own AI gadgets! 👏 If you have any questions, don't hesitate to leave a comment in the comment section at any time! ❤️