Pony Diffusion is a model capable of generating pony, furry, humanoid and other non-photorealistic subjects with high quality, also it has knowledge of various styles. One of the best part of this type models is they have capability of generating almost everything because of their wide-range training, with creativity and correct prompting users can achieve their desired results quickly.

SeaArt - Model Page

What This Model Excels At

• Has Pony Diffusion based special tags for better experience.
• Generating Pony/Furry(feral-anthro)/Humanoid/Human and other species.
• Generating female/male/transgender subjects.
• Crafting SFW/NSFW content imagery.
• Crafting images with wide-range themes. (fantasy, sci-fi, medieval etc.)
• Crafting images with various art styles. (pixel art, cgi, line-art etc.)
• Capability of creating non-exist species with users detailed description.
• Recognizes many of Danbooru based tags, but model training is not Danbooru based.
• Knowledge of many well-known anime/cartoon/game characters.
• Knowledge of various face/body/characteristic/outfit terms.
• Can create amazing background/landscapes.

Cautions For Optimal Usage

• Trained to work without `negative prompt`
• Does not require quality modifiers like `masterpiece, best quality, hd` etc.
• Make sure to use Clip Skip at `2`.
• Can't create photorealistic images.
• Use special data selection tags to achieve desired results.
• `score` tags must be used front of prompt.
• Can create "pseudo signatures" in images because of an training issue, they can be hard to remove even with negatives.
• Use negatives just for unwanted stuff or style, e.g `hat, yellow, 3d, kid`
• 1024px resolution is recommended, although model can generally work with most of supported SDXL resolutions.

Recommended Settings

• Steps: 20-40 (25)
• CFG Scale: 7-8.5
• Sampler: Euler a (can work great with most of Samplers)
• Clip Skip: 2
• VAE: sdxl_vae

Suggested Resources

LoRA

• Style LoRAs for Pony Diffusion V6 XL

Prompt Style

• Requires `score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up,` in front of positive prompt.

Note:`score_9` is much weaker than full string because of an training issue. Prefer suggested full string for better results.

• Does not need any negative or `masterpiece, best quality` like terms.
• Data selection tags can be used to make AI focus on certain contents.
• Can work with both of short and long prompts.

Recommended Prompt Order

1. `score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up`
2. Data Selection Tags (User Preference)
3. Describe what you want in general order.
4. Tags for better understanding and great results.

Special Data Selection Tags

• Tags can be used in both positive and negative prompts to focus on a specific data of the Model's training database.

Source

`source_pony`

`source_furry`

`source_cartoon`

`source_anime`

Content Safety Rating

`rating_safe`

`rating_questionable`

`rating_explicit`

Prompt Examples

Cute Friend

Positive Prompt

`score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_safe, beautiful girl, smile, tongue out, blue eyes, purple hair, messy bun, v, sitting on chair, looking at viewer, classroom, sailor uniform`

Negative Prompt

`3d, greyscale, child, childish`

Monkie Pie

Positive Prompt

`score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_safe, source_furry, source_pony, solo, gigantic, gorilla pony, pinkie pie, pink fur, muscular, gorilla arms, gorilla legs, curly tail, cute face, city`

Negative Prompt

`greyscale, kid, cgi, 3d`

You can join SeaArt Discord Community for more content, also you will have opportunity to choose next model guides with your vote. You can use comments or SeaArt server #guides-feedback to leave feedback or suggestion. We look forward to seeing you on the server.

Happy Generations To Everyone!