Anima
Anima is an open-weight anime and illustration model from CircleStone Labs, built for clean linework, expressive characters, and vivid color straight from a prompt. It reads booru-style tags and natural language, and pairs with a fast-growing library of community LoRAs. Generate with it right here on Civitai — no GPU, no install.
About Anima
Anima is a 2-billion-parameter text-to-image model built through a collaboration between CircleStone Labs and Comfy Org. It is focused on anime concepts, characters, and styles, and extends to other non-photorealistic, illustrative art — but it is explicitly not built for photorealism and will not do it well. It was trained on several million anime images plus roughly 800k non-anime artistic images, with no synthetic data, and an anime knowledge cut-off of September 2025. Architecturally it is compact and Comfy-native: it pairs the diffusion model with a Qwen 3 0.6B text encoder and the Qwen-Image VAE, and reads Danbooru-style tags, natural-language captions, or any mix of the two.
Anima ships as a small family rather than a single checkpoint. Anima-Base is the pretrained, unrefined model — maximum flexibility, diversity, and style adherence, and the version you should train LoRAs against. Anima-Aesthetic is fine-tuned on high-quality images for a more consistent, better default art style, and no longer needs quality tags in the prompt. Anima-Turbo is a distilled few-step variant meant for CFG 1 and roughly 8–12 steps: it generates fast and gains stability and a strong default style, at the cost of some diversity. The authors suggest starting with Turbo for quick iteration, since it is only slightly behind Aesthetic while being much cheaper on step-scaled platforms.
Choose Anima when you want clean anime and illustration work from a lightweight model — its Danbooru vocabulary, artist-tag control, and low compute footprint make it fast and cheap to iterate on. Because it runs at native ~1MP resolutions and is a preview checkpoint, it is strongest at square-ish and standard portrait/landscape sizes rather than very large canvases, and its plain base style rewards adding quality and artist tags. For heavy character-and-booru breadth or the deepest LoRA libraries, the SDXL-based Illustrious, Pony, and NoobAI ecosystems still lead — but Anima trades that scale for speed, a modern September-2025 anime knowledge base, and a growing set of community LoRAs.
How to prompt Anima
- Prompt with Danbooru-style tags, natural-language captions, or a mix of both — Anima was trained with tag dropout, so you do not need to list every relevant tag. When using tags, follow the order [quality/meta/year/safety] [1girl/1boy/etc] [character] [series] [artist] [general tags], lowercase with spaces instead of underscores.
- Prefix every artist tag with "@" (e.g. "@nnn yryr"). Without the "@" the artist effect is very weak. Artist and quality tags meaningfully improve aesthetics because the base checkpoint is intentionally neutral — though Anima-Aesthetic already looks good and does not need quality or score tags.
- Prompt weighting works, but Anima needs a higher weight than SDXL to register — reach for something like (word:2) rather than the usual (word:1.2). Beyond emphasis, steer mainly with tags and description.
- Negative prompts are supported and useful, especially for quality ("worst quality, low quality, jpeg artifacts") and for safety steering — add a safety tag (safe / sensitive / nsfw / explicit) in the positive and/or negative to control content.
- When naming a character, also describe their basic appearance (hair, eyes, outfit) — critical in multi-character scenes, where names alone make the model confuse characters. If prompting in pure natural language, write at least two sentences; very short prompts give unpredictable results on this preview checkpoint, and you can place quality/safety/@artist tags at the start of an NL prompt.
Featured Anima models
Curated — the models worth generating with first.
Popular Anima LoRAs & add-ons
Top LoRAs by downloads — live data, refreshed daily. Stack them on any checkpoint.
![[Anima & SDXL] Artist Style: ashima/アシマ — Anima LoRA preview](https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a0b53e70-abcd-40cf-920a-fb0e63fa0cf7/anim=false,width=450,fit=cover,optimized=true/a0b53e70-abcd-40cf-920a-fb0e63fa0cf7.jpeg)
Example generations
Curated, safe-for-work showcase — every image ships with its prompt and settings.
How to run Anima
Two paths — one takes ten seconds, one takes an afternoon.
⚡ Run on Civitai (Recommended)
The fastest way to start.
🖥️ Run it locally
For power users who want full control.
No graphics card? The Civitai path above skips all of this.
Anima vs other ecosystems
Backed by Civitai usage data.
| Feature | Anima | Illustrious | Pony | NoobAI |
|---|---|---|---|---|
| Best for | Anime & illustration | Anime & illustration | Stylized anime art | Anime, booru-accurate |
| Prompt adherence | Very good | Very good | Good | Very good |
| Text in images | Fair | Fair | Weak | Fair |
| Speed on Civitai | Fast (3–6s) | Medium (5–10s) | Fast (3–6s) | Medium (5–10s) |
| LoRA ecosystem | Growing | 294,000+ | 149,000+ | 9,800+ |
| Available on Civitai | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
Frequently asked questions
What is Anima?
How much does it cost to generate with Anima?
How do I prompt Anima?
Can I train or use Anima LoRAs?
Can I use Anima commercially?
Start generating with Anima now
No installation. No GPU. Runs in your browser.