Anima

Anima is an anime-focused architecture. CoCo’s trial model, Anima Aesthetic v1.1, is an Anima model.

Versions

The official Base, Aesthetic and Turbo versions all work, as do community fine-tunes of Anima. For the Aesthetic version, the official recommendation is not to use score tags.

Required files

File	Goes in	File name
Main model	`diffusion_models`	for example `anima-aesthetic-v1.1`
VAE	`vae`	`qwen_image_vae`
Text encoder	`text_encoders`	`qwen_3_06b_base`

The VAE and text encoder can be downloaded in ⚙ Settings → Resource folders → Download models. When you download the trial model there, missing files are downloaded with it.

Recommended settings

Item	Default
Steps	30
CFG	4
Sampler	ER-SDE
shift	3.0

The samplers grouped under “Anima, Z-Image, Krea 2 & Qwen Image only” in the sampler list suit these architectures. A higher shift changes the composition more.

Quality tags

Anima’s default quality tags and default negative prompt (following the official recommendation):

masterpiece, best quality, score_7

worst quality, low quality, score_1, score_2, score_3, artist name, blurry, jpeg artifacts, chromatic aberration

When the Aesthetic trial model is downloaded through CoCo, it gets a version without score tags.

Graphics memory

“Balanced” under More → VRAM in the generate panel suits most situations: it uses much less graphics memory than “Speed” and is only a little slower. “Low” is only needed when graphics memory is short (for example 8 GB or less).

LoRA and training

Anima can only use Anima LoRAs; SDXL LoRAs are marked incompatible and skipped.
CoCo can train Anima LoRAs; see Training LoRA.

Licence

Anima is licensed by CircleStone Labs under the CircleStone Labs Non-Commercial License: the model itself may not be used commercially; generated images are unrestricted. See the model page for the full terms.