Animagine XL 4.0

Animagine XL 4.0, from Cagliostro Research Lab (initial release January 24, 2025; 4.0 Opt and 4.0 Zero variants added February 13, 2025), is an anime-specialized finetune of Stable Diffusion XL 1.0. It was trained with tag-based captions and a strict tag-ordering method, so it responds best to structured prompts: subject, character name, series, rating, then quality tags like masterpiece and high score. The 4.0 update focused on stability, anatomy, and color accuracy. It is one of the few anime models with thorough official prompting documentation covering quality tags, score tags, negative prompts, and per-orientation resolutions.

The weights carry a CreativeML Open RAIL++-M license and run in ComfyUI, Stable Diffusion WebUI, and diffusers, with a Replicate deployment under aisha-ai-official for API access. Recommended settings are modest, CFG 4-7, 28 steps, Euler Ancestral, and the documented resolutions run up to 640x1536 portrait. It has no published benchmark scores, and like all tag-trained anime models it rewards learning its dialect; natural-language-only prompts work less well than the structured tag format. For clean anime illustration without the explicit-content training mix of the Pony line, it is the most documented option in its class.

No published benchmark scores