Pony

Pony Diffusion V6 XL is an SDXL fine-tune by AstraliteHeart, built as a versatile character and stylized-art model that handles SFW and NSFW output across anthro, feral, and humanoid subjects. Its defining feature is an opinionated "score_" prompt template: prefixing a prompt with a quality chain steers the model toward higher-fidelity results with no negative prompt and otherwise default settings. Because it was trained on a mix of natural-language captions and booru-style tags, it reads plain descriptions and comma-separated tags equally well, and it recognizes a wide range of popular and obscure characters and series.

The score system is the model's signature lever. V6 XL expects the full chain — "score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up" — which the author notes was itself a training quirk that arrived too late to correct; the shorter "score_9" from earlier Pony versions still works but has a much weaker effect. Beyond quality, it exposes special data-selection tags: "source_pony", "source_furry", "source_cartoon", and "source_anime" to bias overall style, plus "rating_safe", "rating_questionable", and "rating_explicit" to control content. It must be loaded with clip skip 2, or generations degrade into low-quality blobs.

As an SDXL fine-tune, Pony inherits SDXL's native resolutions and is compatible with the enormous SDXL-family LoRA library, and it anchors one of the largest community and LoRA ecosystems on Civitai. Choose it when you want deep character knowledge, the tag-driven "score_" workflow, and that vast LoRA depth. For cleaner anime linework and newer concepts, the sibling SDXL fine-tunes Illustrious and NoobAI are strong alternatives; reach for base SDXL when you want a neutral, general-purpose starting point instead of Pony's opinionated defaults.