Original title: 0ppxnhximxr/snapref-v2a
A community author created this character reference adapter for the anime base model Anima base v1.0: provide one character image, and in the prompt specify only the scene, pose, and lighting; hairstyle, eye color, and clothing are inherited from the reference image. There are two modes: Character & Style (cs) inherits the art style from the reference image as well, while Character (char) extracts only the character, with the art style determined by the prompt. The simple node accepts up to 8 images of the same character, for example, one clear face shot plus one full-body shot; the advanced node allows setting individual weights for each reference image. The author compared v1a across a fixed set of 64 use cases × 2 seeds, finding that metrics were largely comparable, but subjectively noted significantly better adherence to the reference.
Why it matters When creating multi-scene illustrations of the same character, you don't need to train a character LoRA first; you can stack multiple reference images and adjust their weights individually
Rising score 48 · Likes 50 · Downloads 1740
Reasons
- Includes six ComfyUI workflows: single reference, character mode, multi-reference, reference weighting, and side-by-side dual characters
- Weights are approximately 382 MB, with an additional FP8 version of about 195 MB that is restored to BF16 during inference
Usage
- First, install Anima base v1.0 along with its text encoder and VAE in ComfyUI; copy the repository's comfyui/snapref/ directory to ComfyUI/custom_nodes/snapref/ and restart
- Place snapref_v2a.safetensors into models/loras/, open workflows/snapref_v2a_simple.json, and select reference images in the SnapRef node; the author's example uses CFG 4.5, Euler / simple, and Fidelity 1.0
- You must use SnapRef's own nodes; do not load it using LoraLoader or generic diffusers pipelines
Risks
- The weights are derived from CircleStone models and are released under the CircleStone Labs Non-Commercial License v1.2, meaning they are for non-commercial use only
- Limitations listed by the author: small accessories, collars, and fine patterns may change; facial details are difficult to reproduce when the face in the reference image is very bright; text or signatures in the reference image may be copied
- This entry has not been run; refer to the model card for actual performance