Skip to main content

​
Anima Prompting Guide

Anima is a text-to-image model built for anime, illustration, and non-photorealistic art. It can make polished character art, stylized scenes, and painterly images, but it is not designed for realistic photos. Use this guide when you want cleaner anime prompts, stronger character control, and fewer accidental style or safety misses.

​
Best Starting Point

Start with this positive prompt prefix: 
masterpiece, best quality, score_7, safe,
Use this negative prompt: 
worst quality, low quality, score_1, score_2, score_3, artist name
Then add your subject, character, style, and scene details after the prefix.

​
Tag Prompt Formula

For tag-based prompts, use this order: 
[quality/meta/year/safety tags], [1girl/1boy/1other], [character], [series], [artist], [general tags]
Example: 
masterpiece, best quality, score_7, safe, 1girl, original character, newest, highres, long black hair, blue eyes, school uniform, cherry blossoms, soft smile, looking at viewer, sunset, warm lighting, detailed background
The sections do not need to be perfect, but keeping the big pieces in this order helps Anima understand what matters most.

​
Useful Tag Groups

Quality tags
  • masterpiece
  • best quality
  • good quality
  • normal quality
  • low quality
  • worst quality
Score tags
  • score_9
  • score_8
  • score_7
  • score_6
  • score_5
You can use quality tags, score tags, both, or neither. A reliable default is masterpiece, best quality, score_7. Time period tags
  • newest
  • recent
  • mid
  • early
  • old
  • year 2025
  • year 2024
Use these when you want the image to feel closer to a specific era of anime art. Meta tags
  • highres
  • absurdres
  • anime screenshot
  • official art
  • jpeg artifacts
Safety tags
  • safe
  • sensitive
  • nsfw
  • explicit
Use safe in the positive prompt when you want non-explicit output. Add unwanted safety classes to the negative prompt if the image keeps drifting.

​
Tag Hygiene

  • Use lowercase for tags.
  • Use spaces instead of underscores, except score tags like score_7.
  • When a tag name differs between Danbooru and Gelbooru, use the Gelbooru-style tag.
  • Do not list every possible detail. Anima was trained with tag dropout, so a focused set of important tags is usually better than a cluttered prompt.

​
Artist Tags

Prefix artist names with @. 
@artist name
The @ matters. Without it, the artist reference may be much weaker. Example: 
masterpiece, best quality, score_7, safe, 1girl, original character, @artist name, silver hair, red scarf, winter street, snow, cinematic lighting
Only use artist references when you have the right to do so and the result is appropriate for your project.

​
Natural Language Prompts

Anima also understands normal descriptive prompts. Use at least two sentences when writing pure natural language; very short prompts can produce unpredictable results. Example: 
masterpiece, best quality, score_7, safe. Digital anime illustration of an original young mage with short auburn hair and green eyes, wearing a navy cloak with silver embroidery. She stands in a moonlit library, holding a glowing spellbook while dust motes float through the air. Warm candlelight, detailed background, gentle expression, crisp line art.
For named characters, describe their appearance after the name. This is especially important when prompting multiple characters. 
Digital anime illustration of Character A, a tall girl with long purple hair and violet eyes, standing beside Character B, a short boy with messy brown hair and round glasses. They are arguing in a crowded train station at night, neon signs reflected on the wet floor.

​
Mixing Tags And Sentences

You can combine tag blocks and prose in the same prompt. A good structure is: 
[quality tags], [safety tag], [character count], [core visual tags]. [One or two natural language sentences with the scene, action, lighting, and mood.]
Example: 
masterpiece, best quality, score_7, safe, 1girl, original character, highres, blue hair, raincoat, umbrella. A quiet anime street scene at dusk, with the girl standing under a red umbrella while rain ripples across puddles. Soft reflections, melancholy mood, detailed storefronts, cinematic composition.

​
Non-Anime Art Prompts

Anima can also respond to non-anime illustration captions. For these, use a dataset-style label at the beginning, then describe the artwork in natural language. 
deviantart
Moonlit Garden
Digital painting of a quiet fantasy garden under blue moonlight, with glowing flowers, stone arches, and a small fountain. Soft brushwork, cool colors, dreamlike atmosphere.

ye-pop
Retro Travel Poster
A bold graphic poster of a mountain railway crossing a red bridge above a pine forest. Clean shapes, limited color palette, vintage print texture.
Use this when you want painterly, poster, abstract, or illustration styles that do not need a standard anime tag structure.

​
Prompt Weighting

Prompt weighting works, but Anima may need stronger weights than some SDXL workflows. 
(chibi:2)
Use weighting for a small number of important traits. If everything is weighted, nothing is really prioritized.

​
Generation Settings

The Anima model card recommends:
  • Resolution: between 512 x 512 and 1536 x 1536
  • Steps: 30-50
  • CFG: 4-5
Sampler behavior can change the look:
  • er_sde: neutral style, flat colors, sharp lines; a good default.
  • euler_a: softer and thinner lines; can lean toward a 2.5D look.
  • dpmpp_2m_sde_gpu: similar to er_sde, but can produce more variety and more unexpected results.
If your image looks overcooked, lower CFG or simplify the prompt. If it looks too plain, add stronger quality, style, lighting, and composition tags.

​
Starter Recipes

Clean character portrait 
masterpiece, best quality, score_7, safe, 1girl, original character, highres, short pink hair, amber eyes, white blouse, black ribbon, soft smile, looking at viewer, simple background, studio lighting, crisp line art
Detailed fantasy scene 
masterpiece, best quality, score_8, safe, 1girl, original character, highres, fantasy armor, silver hair, glowing sword. She stands on a broken bridge above a misty canyon as ancient ruins float in the distance. Dramatic clouds, rim lighting, detailed background, painterly anime style.
Anime screenshot feel 
best quality, score_7, safe, anime screenshot, 1boy, original character, messy black hair, school blazer. A tense hallway scene after class, the boy pauses beside a window as orange sunset light cuts across the floor. Subtle expression, cinematic framing, shallow depth of field.
Stylized poster 
masterpiece, best quality, score_8, safe, official art, 1girl, original character, twin tails, futuristic idol outfit, microphone, neon stage, spotlight, dynamic pose, vivid colors, clean graphic composition, poster art

​
Fix Common Problems

The result is too plain
  • Add masterpiece, best quality, and a higher score tag.
  • Add a clear style target such as anime screenshot, official art, painterly anime style, or crisp line art.
  • Describe lighting, composition, and background instead of only naming the character.
The model changes the character
  • Put the character name earlier in the prompt.
  • Describe hair, eyes, outfit, body type, and signature accessories.
  • For multiple characters, describe each character separately.
The image drifts into unwanted content
  • Keep safe in the positive prompt.
  • Put unwanted safety categories in the negative prompt.
  • Make the scene more specific so the model has less room to improvise.
Text in the image looks wrong
  • Avoid long written phrases inside the image.
  • Ask for a simple sign, logo, or single word only when needed.
  • Add the exact wording afterward with an editor if the text must be accurate.

​
Final Checklist

  • Start with masterpiece, best quality, score_7, safe.
  • Put quality, year, meta, and safety tags before character and scene tags.
  • Use lowercase tags with spaces instead of underscores, except score tags like score_7.
  • Prefer focused tags over exhaustive tag lists.
  • Prefix artist tags with @.
  • Write at least two sentences for pure natural language prompts.
  • Use a detailed negative prompt when the output drifts.
  • Avoid realism prompts; Anima is strongest at anime, illustration, and stylized art.
Reference: circlestone-labs/Anima on Hugging Face