EverAnimate Extension Chunk

The whole reason anyone touches this pack is to get past Wan's ~81-frame native limit, and EverAnimate Extension Chunk is the node that does the actual extending. Where the Initial Chunk starts the video, each Extension Chunk samples the next window, carries the motion forward, and hands you a bigger stack of images to keep chaining. You use one per additional chunk; want a 5-chunk video, you chain four of these after the initial one.

Its inputs are settings (from the Initial Chunk or the previous Extension Chunk), the previous cumulative images, and three carry-control knobs behind a use custom settings toggle. Its outputs mirror the Initial Chunk: settings for the next chunk and images for the next one (or the Video Combine at the end).

What the three knobs actually do

• num_motion_latents (default 1) - how many previous latent frames become motion memory. One latent is the recommended starting point; more buys temporal memory at the cost of being locked to the previous chunk's look.
• continue_motion_max_frames (default 5) - how many of the previous chunk's decoded RGB frames get carried into the next window. This is native WanAnimate's image carry-over, and it's the mechanism doing most of the seam-smoothing.
• motion_handoff_strength (default 0.75) - how hard the next chunk locks onto what it was handed. At 1.0 it's the original hard handoff, which can read as a hard cut; the README and tooltips both point you at 0.6–0.8 for low-step distilled runs where a softer lock looks smoother.

Internally the node keeps a latent of motion memory in the settings socket and uses WanAnimate's returned video_frame_offset to backtrack, so pose, face, background, and mask guides stay aligned with the frames even after carry frames eat into the window. That alignment bookkeeping is exactly the part that's miserable to do by hand, and exactly what this node hides from you.

The one failure mode you'll actually hit

The images input is required, and the node raises a loud error if you leave it unconnected: "EverAnimate Extension Chunk needs the previous chunk images connected to its images input." This is the node's most common support question, and the answer is always the same - you skipped a link in the chain. Trace it: Initial Chunk images → first Extension Chunk images, and so on down the line.

Install

ComfyUI Manager, search "Comfy EverAnimate", or:

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

Restart. No dependencies or model downloads here - the native WanAnimate stack (Animate 14B, umt5_xxl, Wan 2.1 VAE) plus the Master and Initial Chunk nodes upstream are what you actually need.

Common issues

• Seams still visible at joins. Drop motion_handoff_strength toward 0.6 and confirm continue_motion_max_frames is actually carrying (5 is the default). The boundary Color Correction node in this pack exists precisely for the residue this leaves behind.
• Guides drifting out of sync. Make sure each chunk chain carries the offset forward; the Extension node handles it internally, but only if every chunk comes from the same settings lineage.
• use custom settings off and your typed values do nothing. Same pattern as the Initial Chunk - the toggle gates the widgets. Off means defaults.
• Keep the seed shared across every chunk. Different seeds per chunk is the fastest way to a character that swaps identity at every boundary.