ComfyUI Adds Six H3 Max Nodes: Recast to Lip Sync

Introduction
The H3 Max editing tools just moved into ComfyUI! ComfyUI merged six new MiniMax H3 Max partner nodes into core on October 31 at 08:32 UTC (PR #17424): Recast, Extend Video, Reframe Video, Insert Video, Lip Sync and Camera Control.
Until now these H3 Max tricks lived behind separate API endpoints. Now they're ordinary nodes under partner/video/MiniMax, so you can chain them with your loaders, upscalers and editors in one graph. They're closed, API-backed nodes billed through your Comfy account (sign in or use a Comfy API key), not local weights. The same PR also teaches the existing H3 Reference to Video node to read @Image1, @Video1 and @Audio1 tags in your prompt.
The six new nodes
MiniMax H3 Max Recast
Swap the people in a clip for the people in your reference photos. It keeps the motion, camera, cuts and sound.
- Source: 5 to 30 seconds, with no single shot longer than 15 seconds
- References: 1 to 4 photos, at least 256×256. By default they replace the main people from left to right
- Prompt: optional. Say who becomes whom with tags, like "the man becomes @Image2"
- Options: 768P or 1080P, plus an enhance realism toggle for more natural skin and blending
- Price badge: about $0.43 per second at 768P and $0.64 at 1080P
MiniMax H3 Max Extend Video
Continue a clip with new footage guided by a prompt, and up to three optional audio references for voice or sound.
- Source: 1.625 to 60 seconds, up to 50 MB
- New footage: 1 to 15 seconds, at 480P, 768P, 1080P or 2K
- Output modes: extended returns the whole video re-rendered with the new part. continuation returns only the new footage, starting with a 1.6 second overlap
- Models: H3 Max or the half-price H3 Max Turbo
- Price badge: per second of new footage, $0.07 at 480P, $0.11 at 768P, $0.23 at 1080P, $0.46 at 2K (Turbo is half)
MiniMax H3 Max Reframe Video
Turn a 16:9 shot into 9:16 (or 1:1, 4:3, 3:4, 21:9, 9:21) by putting it on a new canvas and generating the missing edges. Timing and sound stay put.
- Source: 0.5 to 300 seconds
- Resolution: 480P, 768P or 1080P, with an optional prompt for the new surroundings
- Price badge: roughly $0.07 to $0.23 per second depending on resolution
MiniMax H3 Max Insert Video
Cut out a stretch of your clip and drop in a brand-new scene, then go back to the original.
- Timing: set a start_time and a resume_time. The new scene runs 5 to 13 seconds
- References: up to 9 images (@Image1 to @Image9) and up to 3 videos (@Video1 to @Video3, 15 seconds total)
- Extras: color match to the source, 480P or 768P
- Tip from the tooltip: references without a prompt are often ignored, so always describe the new scene
MiniMax H3 Max Lip Sync
Animate a still image so its mouth follows an audio track. The video lasts as long as the audio.
- Audio: 5 to 899 seconds, so a full 15-minute monologue fits
- Resolution: 480P, 768P, 1080P or 2K, with an optional transcription pass to guide the sync
- Price badge: from about $0.07 per second at 480P to $0.46 at 2K
MiniMax H3 Max Camera Control
Start from one image and fly the camera to a new angle and distance around the subject.
- Controls: azimuth (−360 to 360°), elevation (−90 to 90°) and distance (0.1 to 10, where 1 keeps the framing)
- Plug-in camera: connect a Create Camera Info or Compose Camera Angle Prompt node and it sets the angle for you
- Frozen scene: leave the prompt empty and only the camera moves, which is lovely for product orbits
- Length: 1 to 15 seconds at 480P, 768P or 1080P
Heads-up from the node itself: azimuth is a target, so a 90° orbit often lands between 30° and 90°.
How to get them
The nodes are on ComfyUI's master branch today, so they show up in nightly builds right away and should reach the regular release and Desktop with the next update. If you run from git:
cd ComfyUI
git pull
pip install -r requirements.txtRestart ComfyUI, sign in to your Comfy account, then search the node menu for MiniMax H3 Max. Each node shows a price badge before you hit Run.
A prompt to try with Recast
Wire two reference photos into image_1 and image_2, then paste this into the Recast prompt:
The woman on the left becomes @Image1 and the man on the right becomes @Image2. Keep their outfits, the café background, the camera move and the original dialogue.Why it matters
H3 Max already had great editing tools, but each one was its own API call. Having all six as nodes means one graph can recast a scene, reframe it to vertical, insert a new beat and lip-sync a voiceover, with no glue scripts. For anyone cutting social versions of a hero video, that's a real time-saver.
Conclusion
This is one of the biggest partner-node drops of the month. If you've been juggling H3 Max endpoints, try building one end-to-end graph this weekend and see how far you get.
Want to try H3 without wiring a graph? Try this on Gen → https://artrealmai.com/gen?utm_source=magazine&utm_campaign=gen&utm_content=comfyui-h3-max-partner-nodes-recast-extend-insert-lip-sync
Original Source
- ComfyUI PR #17424: H3 Max Recast, Extend, Reframe, Insert, Lip Sync and Camera Control nodes
- Merged commit 7c4390a
- nodes_minimax.py at that commit
Happy editing, and may every take land on the first try!
—Aurelia ♡
