An AnimateDiff-style motion module for Anima, with a ComfyUI node and a "freeze or animate anything" workflow.Anima is an image model. This module adds temporal attention after each of its 28 blocks, so 16 frames can be generated together and stay coherent. Anima itself stays frozen, so every frame keeps full Anima image quality.<h3 id="what-it-is-good-at-today">What it is good at today</h3>Pick any part of an image with a word (WeVisOne matting: <code>character</code>, <code>face</code>, <code>sky</code>, <code>sea</code>…) and either:<ul><li>freeze it and let everything else flow: a still character while the sea, moon and sky shift around them, AnimateDiff style;</li><li>animate only it: expressions and small gestures over a background that stays pixel-identical.</li></ul><h3 id="keyframe-modes">Keyframe modes</h3><ul><li>Generate the keyframe with a character sheet (Anima Ref Adapter).</li><li>Generate it from text only (T2V); the keyframe becomes the video's character reference.</li><li>Use your own image; it keeps its aspect ratio and becomes the character reference.</li></ul>Output: 16 frames, 2 s at 8 fps. About 30–45 s per clip on an RTX 5060 Ti 16 GB with Anima Turbo.<h3 id="what's-in-the-files">What's in the files</h3><ul><li><code>anima_motion_r2v_v0.1.safetensors</code>: the motion module (117.5M parameters). Put it in <code>ComfyUI/models/anima_motion/</code>.</li><li><code>anima_motion_v0.1_workflow_and_node.zip</code>: the ComfyUI node (<code>custom_nodes/ComfyUI-Anima-Motion</code>), the workflow and <code>INSTALL.md</code> with every other download.</li></ul><h3 id="requirements">Requirements</h3><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima">Anima</a>: Turbo or Base, the Qwen3 0.6B text encoder and qwen_image_vae.</li><li><a target="_blank" rel="ugc" href="https://github.com/ArcAnima/ComfyUI-Anima-Ref-Adapter">Anima Ref Adapter</a> nodes and the <a target="_blank" rel="ugc" href="https://huggingface.co/ElectronGirl/anima-ref-adapter">v2 LoRA</a>.</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/paralaif/ComfyUI-WeVisOne-VL">ComfyUI-WeVisOne-VL</a> for the selection mask (my port of Tencent WeVisOne).</li></ul><h3 id="recommended-settings">Recommended settings</h3><ul><li>denoise 0.8; 0.4–0.6 for subtle breathing; 1.0 mutates.</li><li>Mask margin 4–8 px with a soft edge when freezing (no halo), 16 px for faces.</li><li>When freezing a character, the background pass runs without the charref and with <code>no humans</code> in the prompt (the workflow does this for you).</li></ul><h3 id="experimental:-please-read">Experimental: please read</h3><ul><li>Trained on only 20 clips (MiniMax H3 image-to-video from Anima keyframes, 384×576, 8 fps), 2000 steps.</li><li>Motion is small: expressions and gestures. Backgrounds morph rather than move physically.</li><li>Without a mask it does not start from your image.</li></ul>Next plans: about 300 clips (characters plus fixed-camera landscapes with ambient motion), keyframe conditioning in training, a bigger cloud run, and 32-frame 16:9 and 9:16 clips.Full write-up with result tables, training scripts and the first I2V module: <a target="_blank" rel="ugc" href="https://huggingface.co/paralaif/anima-motion-module">huggingface.co/paralaif/anima-motion-module</a>License: CircleStone Labs Non-Commercial License, as for Anima.

Anima Motion Module (Experimental): freeze or animate anything

First release. Trained with Anima Ref Adapter v2 active (reference-to-video), 20 clips, 2000 steps, RTX 5090.Includes the freeze/animate workflow with three keyframe modes: character sheet, text only (T2V), and your own image.Recommended: Anima Turbo, 8 steps, CFG 1, euler / simple; video denoise 0.8.

Files with SHA256: 1e419613f61974fc3b596e876948eddaab7dc8431a342e8f94f5cc3ed70bd15c