LanPaint 2.0.0: MiniMax H3 Video and Audio Inpainting

ComfyUI Wikinews

LanPaint 2.0.0 adds MiniMax H3 support with a video and audio inpainting pipeline: paint per-frame masks and audio intervals in one editor, then inpaint both in a single pass.

LanPaint released version 2.0.0 on August 9, adding a dedicated MiniMax H3 video and audio inpainting pipeline. LanPaint is the ComfyUI implementation of the training-free inpainting sampler from the TMLR paper LanPaint: Training-Free Diffusion Inpainting with Asymptotically Exact and Fast Conditional Sampling, and this major version lets you repair H3 video and its joint audio stream in a single pass.

Masked inputMask overlayInpainted result
Masked input (painted in the editor)Mask visible overlayInpainted video and audio

How the AV pipeline works

The new pipeline paints video masks and audio intervals in one editor session, encodes both streams into one nested latent, and runs a single sampler pass:

  1. LanPaint_VideoMaskEditor — pick the video, paint per-frame masks on keyframes (SDF-interpolated in between), and drag audio intervals on the waveform. Mask = 1 regenerates, 0 keeps.
  2. LanPaint_AVEncode — encodes video frames and the audio track into one nested latent with the masks attached.
  3. Sampler — run as usual; the audio stream denoises on its own shifted sigma schedule.
  4. LanPaint_AVDecode — merges the inpainted video with the original at a mask-blended boundary and the inpainted audio inside the masked intervals with a short crossfade, preserving the source fps and bit depth.

You can also export the masks into the video itself: the editor's Export mask video button remuxes a <name>_masked.mp4 copy with the masks embedded in the file metadata, so a masked clip can be reopened later or shared without losing the paint.

How to use it in ComfyUI

Install LanPaint as a regular custom node (ComfyUI Manager or git clone https://github.com/scraed/LanPaint into ComfyUI/custom_nodes/), then download the MiniMax H3 model files and load the official AV inpainting example:

The workflow uses the LanPaint Sampler Custom (Advanced) node. For reference, the README's resource table lists roughly 38–43 GB VRAM for 40–81 frame inpainting at 480p–880p with 20 steps, so a 24 GB card handles shorter clips while high-VRAM setups can go longer.

Availability

LanPaint 2.0.0 is a free open-source ComfyUI extension (GitHub, ComfyUI Registry). The update also includes compatibility fixes for the latest ComfyUI release. Besides MiniMax H3, the sampler continues to support SDXL, Flux, Hunyuan, Qwen-Image, Krea 2, Z-Image, Ideogram 4 and Wan 2.2 inpainting workflows.

Comments

Sign in with GitHub to join the discussion.

Loading comments…
LanPaint 2.0.0: MiniMax H3 Video and Audio Inpainting | ComfyUI Wiki