What is Anima?
Anima is a 2 billion parameter text-to-image model created by CircleStone Labs in collaboration with Comfy Org. It is focused on anime concepts, characters and styles, and also generates a wide variety of other non-photorealistic content. It is a model for illustrations and artistic images, and the vendor is explicit that it will not work well at realism. Training used several million anime images and about 800k non-anime artistic images, with no synthetic data; the anime knowledge cutoff is September 2025.
| Specification | Anima |
|---|---|
| Parameters | 2B |
| Modality | Text to image |
| Text encoder | Qwen3 0.6B base |
| VAE | Qwen-Image VAE |
| Resolutions | 512 to 1536 pixels squared |
| Versions | Base, Aesthetic, Turbo |
| Release date | January 2026 |
| License | CircleStone Labs Non-Commercial |
Anima versions
The repository ships three variants. Anima-Base is the pretrained, unrefined base model with maximum flexibility and diversity, and the version LoRAs should be trained on. Anima-Aesthetic is fine-tuned for better consistency and a higher quality default art style. Anima-Turbo is a distilled version for fast generations at CFG 1 and 8 to 12 steps; CircleStone recommends starting with Turbo, since on average it is only slightly worse than Aesthetic while being much faster to iterate with.
Prompting follows Danbooru conventions: lowercase tags with spaces instead of underscores, artists prefixed with @, and quality tags such as masterpiece and best quality up front. Natural language captions work too, and the two can be mixed in arbitrary order. For Base and Aesthetic the vendor suggests 30 to 50 steps at CFG 4 to 5.
Anima hardware requirements
The download is small for an image model. Each diffusion checkpoint is 4.18 GB, the Qwen3 0.6B text encoder is 1.19 GB, and the Qwen-Image VAE is 0.25 GB, so a full setup is about 5.6 GB on disk. The model is natively supported in ComfyUI and generates between 512 and 1536 pixels squared, which puts a consumer GPU with 8 GB or more of VRAM in comfortable territory.
| File | Goes to | Size |
|---|---|---|
| anima-turbo-v1.0.safetensors | models/diffusion_models | 4.18 GB |
| qwen_3_06b_base.safetensors | models/text_encoders | 1.19 GB |
| qwen_image_vae.safetensors | models/vae | 0.25 GB |
How to run Anima locally
Anima runs natively in ComfyUI. Download the three files from the circlestone-labs/Anima repository, place them in the matching ComfyUI model folders, and open the example workflow image from the repo; ComfyUI loads the whole graph from it. Swap the checkpoint between Base, Aesthetic and Turbo depending on whether you are training LoRAs, chasing the best default style, or iterating fast.
For language models you can run next to it, see the full catalog of models that run locally in Atomic Chat.
Anima license
The weights ship under the CircleStone Labs Non-Commercial License: model use is limited to non-commercial purposes, and commercial use of the model itself requires a separate license from CircleStone Labs. The license claims no ownership of outputs, so the images you generate can be used for any purpose, including commercial work. Read the license file on the repository before deploying it anywhere revenue touches.