Anime character prompts

Anime models speak danbooru tags, not sentences. Here is the tag order that works on Pony and Illustrious, the quality blocks each one wants, and prompts you can open and run.

Best model: Illustrious-XL or Pony Diffusion — both trained on danbooru tags. (Illustrious skips the score ladder; Pony needs it.)

Anime and illustration checkpoints are trained on the danbooru tag vocabulary, so they respond to tags (1girl, solo, cowboy shot, looking at viewer) far better than to natural sentences. Order matters: quality block, then subject count, then appearance, then pose and composition, then setting and lighting.

The big gotcha is per-model quality syntax. Pony wants the score ladder (score_9, score_8_up...); Illustrious does not — give it a short "masterpiece, best quality, very aesthetic" instead. Mixing them up tanks quality.

Example prompts

Hit Open in Builder to drop any of these straight into the workspace, then tweak, weight and generate.

Illustrious — character portrait
masterpiece, best quality, very aesthetic, 1girl, solo, long silver hair, blue eyes, kitsune ears, multiple tails, red kimono, cowboy shot, looking at viewer, cherry blossoms, soft lighting, detailed background

Negative: worst quality, low quality, bad anatomy, extra digits, jpeg artifacts, watermark, signature

Pony — dynamic action
score_9, score_8_up, score_7_up, source_anime, 1girl, solo, white hair, glowing eyes, dynamic pose, motion blur, energy effects, night city rooftop, neon lights, (detailed eyes:1.2)

Negative: score_4, score_5, score_6, worst quality, bad anatomy, extra limbs, fused fingers, blurry

Tips that actually move the needle

  • Decide the model first: Illustrious uses "masterpiece, best quality, very aesthetic"; Pony uses the "score_9, score_8_up, score_7_up" ladder. Do not mix them.
  • Start with subject-count tags (1girl, solo, 2girls) — they anchor the composition before anything else.
  • Use composition tags from the booru vocabulary: cowboy shot, upper body, from above, looking at viewer.
  • On Pony, putting "score_4, score_5, score_6" in the negative prompt pushes quality up noticeably.
  • Weight a single tag with (tag:1.2) when one feature is being ignored — keep it under ~1.3 to avoid distortion.

Frequently asked questions

Should I use sentences or tags for anime models?

Tags. Pony and Illustrious are trained on danbooru-style tags, so "1girl, solo, long hair, school uniform, looking at viewer" works far better than a descriptive sentence.

Do I use score_9 with Illustrious?

No — the score ladder is a Pony thing. Illustrious wants a short quality tag like "masterpiece, best quality, very aesthetic". Using score tags on Illustrious can hurt results.

How do I improve anatomy on anime models?

Keep "bad anatomy, extra digits, fused fingers, extra limbs" in the negative, avoid over-weighting tags above ~1.3, and on Pony add the low score tags to the negative prompt.

← All prompt guides

Build it without the syntax wrangling

Prompt Builder turns weighting, LoRA tags and negative prompts into one-click snippets and keyboard shortcuts — free on web, macOS, Windows and Linux.