WanFunInpaintToVideo (Tiled VAE encode)

WanFunInpaintToVideo (Tiled VAE encode) is the tiled-encode version of ComfyUI's core WanFunInpaintToVideo node, and it's the thinnest node in this pack. The honest headline: if you look at the source, it doesn't have its own logic at all - it just calls the pack's WanFirstLastFrameToVideo node with a start image, an end image, and one CLIP Vision output. That's it. So everything in this article is really about that sibling node, with a tiled VAE encode bolted on.

What it does

Wan Fun Inpaint models extend footage from a start and end frame (with optional CLIP Vision context). This node wires up that conditioning and produces the empty latent for the sampler. The VAE encode of your images runs through vae.encode_tiled() - spatial tiles plus temporal frame-chunks - so the encode phase stays cheap on memory.

Inputs you'd set: positive and negative conditioning from your text prompts, vae, start_image and end_image as your keyframes, optional clip_vision_output, plus width / height / length / batch_size and the four tiling knobs (tile_size / overlap / temporal_size / temporal_overlap). Outputs are the usual positive, negative, latent.

Install

The whole pack installs at once:

cd ComfyUI/custom_nodes
git clone https://github.com/stduhpf/ComfyUI--WanImageToVideoTiled

or via ComfyUI Manager (search "WanImageToVideoTiled"), then restart. No extra dependencies - it borrows the first/last-frame logic from the pack's own node and ComfyUI's core helpers.

The honest caveat

The README leaves this one unchecked - meaning the author hasn't personally tested it, unlike WanImageToVideo, Fun Control, FirstLastFrame, VACE, and the Wan22 latent node. It should work, since it's a thin wrapper around a tested node, but if you're on limited time, reach for WanFirstLastFrameToVideo first and come back here only if you specifically need the Fun Inpaint checkpoint path. Two other reminders that apply everywhere in this pack: tiling only helps the encode stage, and pair the front end with VAE Decode (Tiled) on the way out, because decode is usually the slowest step of a Wan render.