Original title: snapref-v2a-pre2
SnapRef reads the reference image once and reuses it at each sampling step, allowing Anima to generate new poses and scenes for the same character. The prompt only needs to describe the scene; there is no need to specify the character's name or appearance. v2a introduces a char mode: it extracts only the character from the reference image, leaving the art style to be determined by the base model and the prompt. It also supports multiple reference images for the same character.
Why it matters You can maintain consistency for anime characters without training a LoRA; the new version allows you to extract only the character without carrying over the art style from the reference image.
Hugs & Likes 20 · Hugs Rising Score 20 · Hugs Downloads 958
License: Non-commercial · ComfyUI: Mentioned in model card · VRAM: ≈1 GB estimated
Reasons
- The repository includes ComfyUI nodes and is compatible with v1a and v1b
- The author explicitly marks this as a preview version; for daily use, v1a is still recommended
Usage
- Install the ComfyUI nodes from the repository, load the adapter, and connect a reference image; select mode char if you want to change the art style
Risks
- This is a preview version, not an official release
- The license is the circlestone-labs non-commercial license, so commercial use is not permitted