EverAnimate Initial Chunk

Every long WanAnimate video starts the same way: with one first chunk that has no previous chunk to lean on. That's the awkward position EverAnimate Initial Chunk is built for. It takes the bundled settings from the Master Settings node, runs the first window, and - crucially - handles the startup handoff, so your video doesn't begin with a blank gray flash before the character shows up.

The node is deliberately tiny. Its inputs are the initial settings socket from the Master, a use custom settings toggle, and one number: startup_carry_frames, defaulting to 1. Its outputs are a settings socket and an images socket, both of which feed the next node in the chain.

Why it exists

Notice what the Initial Chunk does not have: an images input. Later chunks take the previous cumulative frames as their carry, but chunk one has no previous frames. Instead it uses the reference image itself. With startup_carry_frames set to 1 (the default), the node repeats the reference image once and feeds it in as the image carry-over - the first generated frames are anchored to your character instead of hallucinating a start state. The README calls this "reducing startup flashes," which is exactly the effect you'll see on the timeline: frame one is your reference, not noise.

The use custom settings toggle is the pack's way of protecting beginners from themselves. Flip it on and the startup_carry_frames widget becomes editable; leave it off and the backend uses its own default of 1 regardless of what's typed in the box. That's a slightly odd pattern the first time you hit it - you'll type a value, run, and nothing changes until you tick the toggle - but once you know it, it's predictable.

Where the outputs go

• settings → the next Extension Chunk node's settings input. This socket type is different from the Master's initial settings, so you can't accidentally cross-wire them.
• images → your first view of progress (any Preview Image or Video Combine node), and to the first Extension Chunk's images input if you're chaining.

The chunk nodes don't render previews internally, so if you want to watch a long run progress, hook the images output to a Preview node and let the chain update as each chunk lands.

Install

ComfyUI Manager → search "Comfy EverAnimate", or clone manually:

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

Restart ComfyUI. No dependencies, no model files in the repo. You need the native WanAnimate stack - Animate 14B checkpoint, umt5_xxl text encoder, Wan 2.1 VAE - plus the Master Settings node feeding this one.

Common issues

• The node needs its initial settings socket. Connect Master Settings first; there's no way to construct settings standalone, which is intentional.
• startup_carry_frames seems ignored. Flip use custom settings on. Off means "author's defaults," period.
• First frame still flickers? Raise startup_carry_frames a couple of frames and keep the seed shared with every other chunk. If the seed differs per chunk, boundaries fight each other.
• This pack is weeks old and targets the newer comfy_api node API - update ComfyUI if the node won't register.