KurtC PhotoEd

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.

IntroductionI’ve seen a lot of talk about Guidance under Flux and how it’s nothing like CFG - and in fact that CFG itself should always be set to 1 for Flux to work properly, but that the guidance should be between 2 and 100.There’s lots of chat and info out there to read - but I like to see stuff in a very visual way.So I have used my AI Tool - which allows you to swap models/stack loras/ upscale and gives you a notes box for tracking what you are doing on each run when you are experimenting (https://tensor.art/template/765125454565789843) to run this prompt - clean with no Lora’s etc:ParametersPrompt 1photograph of a 3D snail with a giant stained-glass geodesic shell. Phototealistic, detailed, HD, full snail in frame, acid wash background.Prompt 1 OutcomeHonestly with this prompt - between 2 and 7.5 I can’t really see a lot of changes - there are some variations but they repeat through the steps, and similar at the high numbers - there seems to be a repeating pattern. I will try a different prompt.For this next batch of High number guidance - I did start again but with all the same exact parameters and Seed number.Prompt 2Create a hyperdetailed photorealistic macro image of a queen of hearts playing card lying on a card table. The queen is composed entirely of multiple tiny hearts. She looks contemporary, beautiful, and modern. She is portrayed as a temptress in a red silk dress surrounded by smokey red hearts and beckoning the viewer to enter the card with her finger emerging from the card in 3D along with red smoke. Add a red poker chip to the table.(Same Parameters as before)Prompt 2 Outcome Prompt 2 Via Creation Tab (Has Fixed CFG of 7 that user cannot change) and Slider goes from 2 - 20 but gives exceptions above 10 (should go to 100.)

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).