How to use ComfyUI 0.44.0's dubbing, lip-sync and camera nodes

The release adds six MiniMax H3 Max nodes for lip-sync, camera moves, recasting and reframing, plus two video upscalers, and every one bills through Comfy credits.

ComfyUI's v0.44.0 release was published on November 4, 2026, and most of what a video creator will reach for is a new set of camera and lip-sync nodes that run on someone else's hardware. The update adds six MiniMax H3 Max partner nodes for lip-sync, camera control, recasting, extending, reframing and inserting footage, plus two video upscalers. None of them run on your graphics card, and none of them are free.

If you have been watching ComfyUI's partner nodes pile up, this is the release where the graph starts to look like a finishing suite. The catch is unchanged: each node is an API call billed to a Comfy account.

What ComfyUI 0.44.0 adds

The release notes list the new partner nodes and name the work behind them. The six MiniMax H3 Max nodes come from pull request #17424: Recast, Extend, Reframe, Insert, Lip Sync and Camera Control. The same release adds a Pruna P-Video Upscale node, 1080p, 2K and 4K targets on BFL's Flux Video Upscale V2, a Sonilo Music 1.5 node, a Wonder 4 model on Topaz Image Enhance, MiniMax-H3 latent-upscaler and Sol-H3-VAE support, and it moves WaveSpeed's FlashVSR video upscaler to Comfy Router. Bundled workflow templates go to v0.11.101.

The notes also list a VEED Dubbing node and Patina texture nodes. Those are not new here. Comfy's own changelog dates both to the October 28 update, v0.43.1, and SLOP TV covered them that week. The release generator lists them again because it compares the new tag against v0.43.0. Treat the MiniMax H3 Max set as the new part of this version.

Before you wire anything

Partner nodes are not local models. Comfy's overview page states they connect to external APIs, need a Comfy account logged in with a credit balance above zero, and that "Partner Nodes require credits for API calls to closed-source models, so they do not support free usage."

Three steps get you running. Update ComfyUI to v0.44.0 or the current nightly build, since the released version sometimes lags behind the code. In Settings, open the User panel and log in with your Comfy account, then open the Credits panel and top up. Add any partner node from the node library and run it the way you run a local one. On a network that is not loopback, log in with an API key, because Comfy restricts account login to 127.0.0.1 or localhost by default.

Dub a clip with VEED Dubbing

The Dubbing node translates the speech in a clip into another language, cloning each speaker's voice and re-syncing their lips to the translation. It shipped in the October 28 update, so treat it as a working tool rather than a preview. Comfy's node page gives two tiers. standard is mono with 85 target languages, and pro is stereo and reaches 101 languages with regional accents such as English (UK) and Spanish (Mexico). The node takes up to 599 seconds of video, and the output is as long as its audio track at 25 frames a second.

To run it, open the Template Library and search for "VEED Dubbing." Load a clip into LoadVideo, pick the tier in model, set target_language, and run. Two rules from the documentation decide whether it works: send footage with one visible speaker facing the camera, and check that the audio actually contains speech. Music, sound effects and product shots give the model nothing to translate.

It bills per second of output. SLOP TV's October 30 report on the API, citing VEED's newsroom, prices Standard at $0.11 a second and Pro at $0.18.

Make a still photo talk with MiniMax H3 Max Lip Sync

This is the new lip-sync node, and for most creators it is the place to start. It animates a still image so the mouth follows an audio track, and the output is exactly as long as the audio you connect. The node documentation lists a single image of at least 256x256 pixels with an aspect ratio between 0.4 and 2.5, audio from 5 to 899 seconds, and a resolution of 480P, 768P, 1080P or 2K. An optional transcription switch tells the model to transcribe the audio first to guide the sync, and the node returns one video.

The cost is simple per second of output. Comfy's pricing page lists the node at 15.09 credits a second at 480P, 24.14 at 768P, 48.28 at 1080P and 96.55 at 2K, with a 1.2x multiplier once the output runs past 15 seconds.

Orbit a still frame with Camera Control

The Camera Control node starts from one image and moves the camera to a new angle and distance around the subject. You set an end position with azimuth, the horizontal orbit in degrees (default 90), elevation (default 0), and distance, where 1 keeps the framing, below 1 pushes in and above 1 pulls back. duration runs 1 to 15 seconds and defaults to 5, and resolution is 480P, 768P or 1080P.

Leave the prompt empty and the scene stays frozen while only the camera moves, which is how you get a clean push-in or orbit from a single generated frame. Comfy's docs warn that the azimuth is a target rather than a promise: a request for 90 degrees often lands somewhere between 30 and 90. The node costs 15.09 credits a second at 480P, 24.14 at 768P and 48.28 at 1080P.

Recast, Extend, Reframe and Insert

Four more nodes round out the H3 Max set. Recast replaces the people in a clip with up to four faces taken from reference photos, keeping the motion, camera, cuts and sound of the source. It takes 5 to 30 seconds of video with every shot 15 seconds or shorter, and you point at the new faces in the prompt as @Image1 through @Image4. It bills at 90.52 credits a second at 768P and 135.78 at 1080P, the most expensive nodes in the set.

Extend, Reframe and Insert change the length, the framing and the contents of a clip. Comfy lists Extend Video from 15.09 credits a second at 480P up to 96.55 at 2K, and it also runs on the faster H3 Max Turbo model at half rate. Reframe Video runs 15.09 to 48.28 credits a second, and Insert Video 15.09 at 480P and 18.10 at 768P.

Finish in 4K with Pruna and BFL

Two upscalers arrive with the release. The release notes list a Pruna P-Video Upscale node, whose built-in description is to upscale a video to 720p, 1080p, 2K or 4K. BFL's Flux Video Upscale V2 gained 1080p, 2K and 4K targets, and Comfy lists its price by output megapixel: 15.09 credits per megapixel per second in creative mode and 10.56 in precise. That works out to 31.24 credits a second for a 1080p output and 125.10 for 4K, both in creative mode.

What each node costs you

Credits per second of video, from Comfy's partner node pricing page:

Node	480P	768P	1080P	2K
MiniMax H3 Max Lip Sync	15.09	24.14	48.28	96.55
MiniMax H3 Max Camera Control	15.09	24.14	48.28	n/a
MiniMax H3 Max Extend Video	15.09	24.14	48.28	96.55
MiniMax H3 Max Reframe Video	15.09	24.14	48.28	n/a
MiniMax H3 Max Insert Video	15.09	18.10	n/a	n/a
MiniMax H3 Max Recast	n/a	90.52	135.78	n/a

Put the rates side by side and the shape is clear. Lip Sync and Camera Control are the cheap end of the new set, Recast is the expensive one. At the roughly 211 credits to the dollar that SLOP TV worked out from Comfy's cloud GPU table on October 29, the 768P Lip Sync rate of 24.14 credits a second is about 11 cents a second, close to the standalone VEED Dubbing API. Generate drafts at 480P and re-run only the takes you keep at 1080P.

Update ComfyUI to v0.44.0 from the download page or run the nightly build, log in with your Comfy account, top up credits in Settings, and try the MiniMax H3 Max Lip Sync node first.