Comfy EverAnimate

Native ComfyUI custom nodes for experimenting with EverAnimate-style latent anchors plus native WanAnimate image carry-over on top of native sampler workflows.

Nodes

• Comfy EverAnimate: native WanAnimateToVideo-style conditioning node with EverAnimate anchor latents, native image carry-over, optional latent motion memory, pose strength, and face strength.
• EverAnimate Master Settings: collects shared model, video, sampler, anchor, pose/face, and guide settings into an initial settings socket.
• EverAnimate Initial Chunk: generates the first chunk from the initial settings and owns the startup carry setting.
• EverAnimate Extension Chunk: generates later chunks from the previous cumulative images, owns the handoff settings, and carries sampler state forward.
• EverAnimate Color Correction: applies native Transfer Color only around chunk boundaries to reduce color/brightness flashes without changing the whole video.
• EverAnimate Chunks Calculator: previews the total number of chunks needed from the Master pose video frame count and chunk length.

Intended Workflow

Comfy EverAnimate -> native KSampler -> TrimVideoLatent -> VAEDecode -> Video Combine

For smoother chunk boundaries, connect the previous chunk's trimmed decoded images into the next Comfy EverAnimate continue_motion input. This uses native WanAnimate-style image carry-over and takes priority over prev_samples.

For chunk 1, you can connect the reference image into continue_motion and set continue_motion_max_frames to 1 to reduce startup flashes. For later chunks, continue_motion_max_frames defaults to 5.

Master + Chunk Workflow

EverAnimate Master Settings -> EverAnimate Initial Chunk -> EverAnimate Extension Chunk -> Video Combine

Connect the Master initial settings output to the Initial Chunk initial settings input. The Initial Chunk then outputs normal settings for Extension Chunk nodes. These two settings sockets are intentionally different types, so the initial settings cannot be plugged into Extension chunks and normal settings cannot be plugged back into Initial.

For longer videos, add more EverAnimate Extension Chunk nodes. Connect each previous chunk settings output to the next chunk settings input, and each previous chunk images output to the next chunk images input. The final Video Combine connects to the last chunk images.

Optional chunk calculation:

EverAnimate Master Settings chunks calculator -> EverAnimate Chunks Calculator

The calculator previews Total chunks needed followed by the bold chunk count inside the node. Its output socket is an INT containing only the chunk count. It rounds partial chunks up and counts the Initial Chunk as chunk 1.

Optional boundary color correction:

Last Extension Chunk images + settings -> EverAnimate Color Correction -> Video Combine

Use the same reference images you would use with native Transfer Color. The correction node defaults to 12 frames before each join, 16 frames after each join, mkl_lab, per_frame, strength 1.0, and only processes frames near each chunk join.

The initial chunk has no image input because it uses the reference image startup carry. Extension chunks use the previous cumulative images plus latent motion memory stored in the settings socket.

The seed is in Master Sampling Settings and is shared by all chunks. Keep it fixed across chunks for the most stable boundaries. Chunk nodes do not encode previews internally; connect any chunk images output to a normal Preview Image or Video Combine node when you want to inspect progress.

Master chunk length is the per-chunk Wan window length, not the total final video length. ref image background is at the top of the video settings, defaults to enabled, and when it is on, the Master ignores connected background_video and character_mask inputs.

The flash-sensitive settings are on the chunk nodes:

• Initial and Extension chunks have a use custom settings toggle. When it is off, the custom rows stay visible but are locked/gray and the backend uses the original defaults. Turn it on to unlock those rows and use your custom values.
• Initial chunk: startup_carry_frames defaults to 1, so the first generated frames are anchored to the reference image instead of starting from an empty handoff.
• Extension chunk: num_motion_latents defaults to 1, continue_motion_max_frames defaults to 5, and motion_handoff_strength defaults to 0.75. This keeps one latent of motion memory and softly carries five decoded frames across the boundary.
• Extension chunks use WanAnimate's returned video_frame_offset internally so pose, face, background, and mask guides stay aligned after carry-frame backtracking.

Defaults

• num_video_anchor_latents: 4
• video_frame_offset: 0
• Master width: 480
• Master height: 832
• Master chunk length: 81
• Master ref image background: true
• Master seed: 42, fixed
• pose_strength: 1.0
• face_strength: 1.0
• Initial use custom settings: false
• Initial startup_carry_frames: 1
• Extension use custom settings: false
• Extension num_motion_latents: 1
• Extension continue_motion_max_frames: 5
• Extension motion_handoff_strength: 0.75
• Color Correction frames_before: 12
• Color Correction frames_after: 16
• Color Correction method: mkl_lab
• Color Correction source_stats: per_frame
• Color Correction strength: 1.0

Install

Clone this repository into your ComfyUI custom_nodes folder, then restart ComfyUI.

git clone https://github.com/younestft/Comfy_EverAnimate.git Comfy-EverAnimate

This node pack is native-ComfyUI focused and does not depend on WanVideoWrapper.

Credits

This node pack builds on ideas and interfaces from these projects:

• vita-epfl/EverAnimate for the original EverAnimate research and implementation.
• comfy-org/comfyui for ComfyUI and the native WanAnimate nodes this pack integrates with.

Thank you to the authors and maintainers of both projects.