Anima's new system prompt lets LLMs blend Danbooru tags and natural language without breaking wildcards

The Anima prompt skill system prompt solves a common pain point for users of anime-style image generation models (Anima, NovelAI, Stable Diffusion). These models accept both Danbooru tags (comma-separated keywords like `1girl, solo, classroom`) and natural language descriptions. However, feeding only tags loses spatial context (e.g., "Where is the subject?"), and feeding only natural language wastes the precise control of tags. Worse, LLMs often arbitrarily expand wildcards (e.g., turning `{standing|sitting}` into "standing or sitting") or delete unrecognized tags. The system prompt redefines the LLM's role: it must never generate images, never rewrite tags, and never expand wildcards. Instead, it reinforces spatial relationships and visual flow, outputting a single English paragraph without markdown or prefatory text.

For example, input `1girl, {standing, sitting}, classroom, desk, {morning, evening}` yields `masterpiece, 1girl, {standing,| sitting}, in the center of a classroom, positioned in front of a desk, with {morning,| evening} lighting implied by the scene context.` Wildcards are preserved exactly (including syntax), and natural language inputs are converted to structured English with inferred Danbooru elements. This makes the output directly copy-pasteable into image generators, saving artists time and frustration. The prompt is designed for users who mix tags and natural language and want consistent, high-quality results without LLM overcomplication.