Pony Diffusion

Pony Diffusion is a community-driven text-to-image diffusion model forked from Stable Diffusion, focused on stylised and non-photoreal character art. It’s geared toward people who want models that recognise characters and a wide range of styles rather than pure photorealism.

V6 XL was trained on about 2.6 million images and uses score-tag prompting (score_9 / score_8_up) plus broad Danbooru and CLIP caption support, which helps with character consistency and style variety. V7 expands the training set to roughly 10 million images, introduces style grouping, improves SFW coverage and includes video data preparation for possible text-to-video work later on.

The model is available to try for free through embedded Hugging Face Spaces playgrounds and doesn’t require signing up. The project is open source and maintained by its community, so development and dataset changes reflect community decisions and contributions.

• Built on Stable Diffusion with community contributions
• V6 XL: ~2.6M images, score-tag prompting, Danbooru/CLIP captions
• V7: ~10M images, style grouping, better SFW coverage, video data prep
• Free demos via Hugging Face Spaces, no signup needed