ComfyUI adds VEED nodes to lip-sync a still photo

The three cloud nodes also re-sync a speaker's mouth in footage you already shot and cut a subject out with an alpha matte.

ComfyUI's newest release folds three VEED video edits into the node canvas, so a solo creator can build a talking head, a re-dub and a background matte without opening a second app. Version 0.42.4 shipped on October 22, 2026, according to the ComfyUI changelog, and adds VEED Fabric 1.0, VEED Lipsync 2.0 and VEED Remove Video Background and Clean Audio as partner nodes.

Now, one caveat: all three are cloud nodes. They run on VEED's servers, need a Comfy account with credits, and download no model files to your machine. The edit moves into ComfyUI's canvas, but the compute does not move onto your GPU.

What the three VEED nodes do

The release puts a talking-photo generator, a lip-sync dub tool and a matte cutter in one node graph, each billed by the second or the frame. Fabric 1.0 animates a photo from an audio track. Lipsync 2.0 re-syncs a speaker to new dialogue in footage you already shot. Remove Video Background returns the frames plus an alpha mask instead of a flattened video, so the subject composites over a new scene or exports as a transparent sequence. The VEED video nodes documentation spells out the inputs and outputs of each one.

How to make a still photo talk with Fabric 1.0

Fabric 1.0 takes one image and one audio track. You load a single clear, front-facing face into the LoadImage node and the speech into LoadAudio. The speech sets both the lip movement and the clip length. The docs say the audio runs up to 300 seconds, and the video comes back exactly as long as the audio.

Resolution is 480p or 720p, with 720p the default. At 16:9 that is about 864x480 for 480p and 1312x736 for 720p. The clip keeps the image's aspect ratio, so a portrait stays a portrait. The docs note it works on photos, illustrations and mascots, which is how you put a voice on a character that was never filmed.

How to re-dub footage you already shot

Lipsync 2.0 keeps the frame you recorded and replaces what the speaker says. Load one front-facing speaker into LoadVideo and the new speech into LoadAudio. The new speech runs up to 600 seconds and sets the output length, and the video keeps its source resolution at 25 fps.

Two limits are worth planning around. If the footage is shorter than the audio, the node plays the clip forward and then backward until the audio ends, so send a clip long enough when the shot has to keep its direction. And both the ComfyUI docs and VEED's own lip-sync guide say one speaker per video, with more than one not supported. A seed control reruns the job, but the docs say results stay non-deterministic.

How to cut a subject out without a green screen

Remove Video Background returns the frames plus a mask that marks what was background, not a flattened clip. Clips run up to 60 seconds. The bundled template turns the mask into alpha for a PNG sequence. For a transparent video, join the image with its alpha and save as WEBM, setting the codec to vp9, which the docs call the only option that writes the alpha channel. Output resolution has no control: standard and green screen return up to 3840x2160, fast returns up to 1920x1080, and larger input is downscaled.

The node also carries toggles for edge refinement and person detection, both on by default, plus a spill-suppression strength for real green screen plates. Speech cleanup runs through a separate pair of nodes, and the Clean Audio documentation says the audio version returns 48 kHz mono and accepts runs up to 1,799 seconds.

What it costs, and what the docs do not say

Each node bills per second or per frame of the clip you send. VEED's pricing guide lists Fabric 1.0 at $0.08 to $0.15 per second of generated video and Lipsync 2.0 at $0.07 per second of generated video. Because a Fabric clip's length follows the audio, the 300-second audio cap is the longest single run the docs describe, and they state no separate output limit beyond it. The docs give no minimum image resolution for Fabric, so that limit is not documented.

The same release added Mirelo SFX 2.0, Krea 3 Image's ultra thinking mode, a price-badge fix on the HeyGen Voice 1.0 node and Templates v0.11.94, per the changelog. A separate deprecation lands sooner: ComfyUI deprecated the Luma Ray 2 nodes in v0.39.0 and says the Ray 2 endpoints shut down on October 24, 2026.

Start from the "VEED Fabric 1.0: Talking Photo" template in the Template Library, or search VEED in the template list after updating ComfyUI.