Download
1 variant available
bf16 SafeTensor
anima-base-v1.0-hand-pose-lllite-v1.0.safetensors
BF16, good balance • 15.22 MB
Verified: 12 hours ago
2
12
License:
AnimaThe Anima Model is licensed by CircleStone Labs LLC. Copyright CircleStone Labs LLC. IN NO EVENT SHALL CIRCLESTONE LABS LLC BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.
Built on NVIDIA Cosmos
BASE-ONLY RELEASE — trained and validated specifically for Anima-Base v1.0. This is not an Anima-family universal model. Anima-Aesthetic, Anima-Turbo, and other derivatives are not validated or supported.
>
> A hand-pose ControlNet-LLLite for Anima-Base v1.0. It takes a standard Easy-DWPose/OpenPose hand-only condition image with 21 keypoints per hand and constrains hand placement, orientation, joint topology, finger count, and gesture structure before sampling. It is not a LoRA and does not use inpainting.
>
> Training data: 4,000 high-quality anime image/keypoint pairs: 3,272 single-hand and 728 two-hand samples.
>
> Install: place the file in ComfyUI/models/controlnet/Anima-LLLite/. Install ComfyUI-Anima-LLLite and use Apply Anima ControlNet-LLLite (sd-scripts) AnimaLLLiteApply_sdscripts).
>
> Recommended starting settings: strength 1.2, start 0.0, end 1.0, preserve_wrapper true; 30–40 steps, CFG 4, er_sde/simple, denoise 1.0. Use a black-background, standard colored hand skeleton with white joints, matching the target canvas aspect ratio. Explicitly describe the intended hand count and gesture in the positive prompt.
>
> Validation: in a fixed six-pose cloud ComfyUI test, exact hand count improved from 16.7% without LLLite to 83.3% with v1.0 at strength 1.2; NME improved from 1.148 to 0.224 and [email protected] from 0.020 to 0.536. Later v1.2/v1.3 candidates did not reliably outperform v1.0 across three independent blind evaluations, so v1.0 remains the stable release.
>
> Limitations: this model does not guarantee perfect anatomy in every seed. Extreme perspective, crossing hands, severe occlusion, complex object interaction, and three or more hands remain difficult. It constrains 21-point structure, not pixel-level nails, skin folds, or occlusion edges.
>
> License and attribution: this is an unofficial modified derivative of Anima-Base v1.0. Anima and derivatives are governed by the CircleStone Labs Non-Commercial License v1.2 and applicable NVIDIA terms. Model weights and derivatives are non-commercial; generated outputs are treated separately by the upstream license. Review the current upstream license before use or redistribution.
>
> Upstream model: https://huggingface.co/circlestone-labs/Anima
> License: https://huggingface.co/circlestone-labs/Anima/blob/main/LICENSE.md
> Required node: https://github.com/kohya-ss/ComfyUI-Anima-LLLite