Anima Reference Adapter for ComfyUI

Character and style from one image. Pose, place and framing from your prompt.

Give Anima a single reference image and a prompt that only describes the scene. The adapter keeps who is drawn (hair, eyes, outfit, accessories) and some of how it is drawn, while pose, background and composition follow the prompt. No character name, artist tag or appearance tags needed, so it also works for original characters.

Every image below used the prompt masterpiece, best quality, score_7, newest, highres, absurdres, safe, <gender>, solo, <scene> with the scene tags shown, the reference image on the left, and Euler 28 steps, cfg 4.5, reference guidance 4.5 + APG, context detailer LoRA 0.3, base slider LoRA 0.6, Anima Safe PAG (block 18, scale 1.3, 0.2-0.9). Seeds were not cherry-picked.

Reference A input image	full body, walking, hands in pockets, city, street, night, rain	upper body, sitting, holding cup, drinking, cafe, indoors, window, sunlight	cowboy shot, looking at viewer, smile, beach, ocean, sunset
portrait, from side, looking afar, cherry blossoms, petals	upper body, reading, holding book, library, bookshelf	full body, standing, snow, forest, winter

Reference B input image	full body, walking, smile, city, street, night, neon lights	upper body, sitting, eating, cake, cafe, indoors	cowboy shot, looking back, smile, beach, ocean, sunset
portrait, closed eyes, smile, cherry blossoms, petals, wind	full body, sitting, on chair, crossed legs, classroom, sunlight	upper body, holding umbrella, rain, street, looking at viewer

These characters were excluded from training. Each row: the reference image, the adapter's result for a different scene, the same scene-only prompt without the adapter (a generic character), and what the base model draws when the prompt also names the character with its appearance tags.

flowchart LR
    R[reference image] -->|letterbox + VAE| RL[reference latent]
    N[noise] --> X[target latent]
    X --> F[[2-frame clip]]
    RL -->|"frame 1: timestep 0, RoPE h shifted by one image height"| F
    F --> D["Anima DiT + adapter LoRA<br/>(LoRA active only with the reference)"]
    P[scene-only prompt] --> D
    D -->|keep frame 0| O[output]

• The reference becomes a second latent frame, fed clean (timestep 0).
• Its RoPE coordinates are shifted past the target grid, so the model cannot copy it pixel-aligned: no pose copying, unlike an aligned temporal reference.
• A rank-32 LoRA (45.9M parameters) learns to read that frame. It was trained on pairs of the same character in different scenes generated by Anima, with the character's name, artist and appearance tags removed from the target prompt.
• Separate text and reference guidance with adaptive projected guidance (APG) keeps colors under control at high reference guidance.

Details, data and failure modes: technical report.

cd ComfyUI/custom_nodes
git clone https://github.com/ElectronGirl/ComfyUI-Anima-Ref-Adapter.git

1. Download anima_ref_adapter_pilot_r32.safetensors into ComfyUI/models/loras/.
2. You also need Anima base v1.0, the Qwen3-0.6B text encoder and the Qwen-Image VAE from circlestone-labs/Anima.
3. Both example workflows use Patch Sage Attention from ComfyUI-KJNodes. Install it, or bypass that node (Ctrl+B) if you don't have SageAttention.
4. Restart ComfyUI and reload the browser tab (Ctrl+F5).
5. Open workflows/anima_ref_adapter.json, load a reference image, write a scene-only prompt, queue.

Workflows: anima_ref_adapter.json (basic) and anima_ref_adapter_pag.json (with Anima Safe PAG).

The node code is Apache-2.0 (LICENSE). The adapter weights are a derivative of the CircleStone Model and are distributed on Hugging Face under the CircleStone Non-Commercial License. This project is not affiliated with or endorsed by CircleStone Labs.

@misc{anima_ref_adapter_2026,
  title  = {Anima Reference Adapter: Single-Image Character and Style Transfer with an Offset-RoPE Reference Frame},
  author = {ElectronGirl},
  year   = {2026},
  url    = {https://github.com/ElectronGirl/ComfyUI-Anima-Ref-Adapter}
}