The PONY Diffusion model is an intriguing AI tool used for generating high-quality images based on tags for prompts. It leverages Danbooru tags, a structured and detailed tagging system, to enhance the accuracy and specificity of the generated images. Understanding how to effectively craft prompts using these tags can significantly improve the results you achieve with the PONY Diffusion model.
### Understanding Danbooru Tags
Danbooru tags are keywords that describe various elements within an image. They are categorized into several groups, such as character traits, clothing, pose, background, and more. By using these tags, you can guide the PONY Diffusion model to create images that closely match your vision.
### Basic Prompt Structure
When creating prompts for the PONY Diffusion model, it’s essential to follow a structured approach. Here’s a basic structure you can use:
1. Scene Quality: Include any Embedding tags, Score, rating and source tags
2. Camera Specification: Define the distance, angle, distortion level, etc.
3. Character Description: Provide an essential description of the main character.
4. General Details: List clothing or objects defining the character or image
5. Scene Description: Include key factors for modifying the general set.
6. Background Description: Describe the ambiance, architecture, weather, etc.
7. Artist Style: Modify the entire scene in accordance with cultural phenomena.
### Crafting Effective Prompts
1. Start with a Clear Concept: Begin with a clear idea of what you want to generate. For example, an office romance scene with two coworkers.
2. Use Specific Tags: Select relevant Danbooru tags to describe the characters, setting, and actions.
3. Combine Tags Thoughtfully: Ensure the tags you combine make sense together and provide a coherent description.
4. Avoid Redundant Tags: Don’t repeat tags within the same prompt. It adds unnecessary complexity.
### Example Prompts
Here are a few examples using some of your most used tags:
Example 1: Romance Kim Possible and Bonnie Rockwaller
score_9, score_8_up, score_7_up, score_6_up, realistic, wide shot, 2girl, aged up, Kim possible,red lipstick, sexy, beautiful, smooth face, seductive smile, huge breasts, slim waist, thick thighs,green lingerie, thigh high stockings, Bonnie rockwaller in white lingerieExample 2: Inspired by Jewelz Blu
score_9, score_8_up, score_7_up, score_6_up, 1girl, solo, full body, candid shot, sitting, reclining, on bed, ankles crossed, looking at camera, naughty smile, relaxed pose, (blue hair:1.4), (hime cut:1), blue eyes, thin eyebrows, blue eyebrows, slight arch, round face shape, well proportioned nose, (full lips:1), (blue lipstick), high cheekbones, flawless skin, large breasts, narrow waist, wide hips, voluptuous thighs, large hoop earrings,long fingernails, blue nails, blue platform high heels, blue subtle makeup, blue taut pencil dress, inside, stage background, studio lighting, soft shadows, soft lightingExample 3: Princess Zelda Inspired Portrait
score_9, score_8_up, score_7_up, score_6_up, 1girl, aged up, Zelda ,red lipstick, sexy, beautiful, smooth face, seductive smile, huge breasts, slim waist, thick thighs,[purple lingerie|gold lingerie|white lingerie], thigh high stockings### Practical Tips
-Danbooru tags: These tags help in guiding your prompt creation. But that does not mean the tags will always work.
-Existing Character: Try using a character built into the model and augment what your looking for. You can find a list of character and tags here. If you are looking for realistic try not including the source tag.
-Defining age can be difficult: Try using In the prompt, "age XX" where XX is the bottom age in years for my desired range (10, 20, 30, etc.) augmented with the following terms
• "infant" for <2 yrs
• "child" for <10 yrs
• "teen" to reinforce "age 10"
• "college age" for upper "age 10" range into low "age 20" range
• "young adult" reinforces "age 30" range into middle "age 40" range
• "middle age" for upper "age 40" range into lower "age 60" range
• "grandmother/grandfather" for "age 60" on up
I'll use similar terms in the negative prompt to refine to a tighter age range.
- Experiment with Variations: Don’t hesitate to tweak your prompts and experiment with different combinations of tags to achieve the best results.
-Tag Weights: Some tags will automatically out weigh others, so use () to increase weight or (tag:1.x) where x is a number 1-9. To decrease weight use [] or (tag:0.x) where x is 9-1.
- Negative Prompts: Use negative prompts to exclude unwanted elements. For example, `negative_prompt: lowres, source_cartoon, source_anime, sketch, drawing`.
###Useful Links
Tag groups- This is a list of wikis that are themselves lists of tags. Oftentimes, a tag is listed in multiple pages, or in multiple sections of the same page, for better navigation.
Fashion Styles -Tags representing clothing and fashion styles.
Attire -Tags representing different types of clothing.
Dresses - Tags for appearance and models of dresses as well as actions.
Sexual Attire - Tags representing clothing used in intimate contexts.
Shoes - Different shoe styles.
Body parts - Different body parts and augmentations
howto:rate - With ratings replace the : with and _ to get the desired effect.
### Conclusion
Prompt engineering for the PONY Diffusion model using Danbooru tags is an art in itself. By understanding and effectively utilizing these tags, you can significantly enhance the quality and relevance of your generated images. Remember to be specific, avoid redundancy, and always experiment with different tag combinations to find what works best for you.
Happy creating!
