ComfyUI>Workflows>LongCat Video Avatar 1.6 | Lip-Synced Videos

LongCat Video Avatar 1.6 | Lip-Synced Videos

Workflow Name: RunComfy/LongCat-Avatar-1.6
Workflow ID: 0000...1567
Animate your reference image with LongCat Avatar 1.6. Add speech for a lip-synced MP4. Create solo presenters. Build two-person dialogues. Assign voices with SAM 3.1 masks. Switch between dedicated canvas groups.

LongCat Video Avatar 1.6 ComfyUI Workflow Workflow

LongCat Video Avatar 1.6 ComfyUI | SAM 3.1 Lip Sync
Want to run this workflow?
  • Fully operational workflows
  • No missing nodes or models
  • No manual setups required
  • Features stunning visuals

LongCat Video Avatar 1.6 ComfyUI Workflow Examples

LongCat Video Avatar 1.6 ComfyUI Workflow#

The LongCat Video Avatar 1.6 ComfyUI Workflow turns a single reference image and one or two voice tracks into a clean, lip‑synced MP4. It ships with two canvas groups: a default two‑speaker dialogue that assigns voices to faces with SAM 3.1 masks, and a single‑speaker monologue for presenters or characters. Creators, educators, and product teams can quickly produce talking avatars with stable identity, natural mouth shapes, and gentle head motion.

Under the hood, the graph segments longer scripts into short overlapping chunks, generates frames for each chunk with LongCat Video Avatar 1.6, trims the overlap, then stitches the segments back together while attaching the full soundtrack. The result is a smooth, voice‑accurate video that respects your prompt and composition.

Key models in Comfyui LongCat Video Avatar 1.6 ComfyUI Workflow#

  • Meituan LongCat Video Avatar 1.6. Core avatar model that turns speech features plus an identity image into lip‑synced frames while preserving appearance. Model card
  • LongCat Video Avatar adapters (DMD and OPSD) packaged for ComfyUI. Optional adapters that align sampling behavior with the released pipeline. Weights
  • Whisper Large V3 encoder. Extracts robust speech features that drive mouth shapes and coarticulation. Model
  • Segment Anything (SAM) family. Provides high‑quality person masks so each voice track animates the correct face in dialogue shots. Repository

How to use Comfyui LongCat Video Avatar 1.6 ComfyUI Workflow#

Overall flow

  • Choose one group to run. The graph splits the audio into overlapping segments, generates frames for each segment with LongCat, decodes them into images, drops the repeated overlap, and concatenates segments into a single video with your full soundtrack.

Two‑Speaker Dialogue · default

  • Load one image showing both people in the frame using Start image. Add two voice tracks, one per speaker, to Voice A and Voice B. The sample project routes the first line to the woman on the left, then the second line to the man on the right.
  • SAM 3.1 nodes detect each person and output two masks. Change the short text prompts that read woman and man to match your photo if needed. Use the mask previews to confirm that the left and right faces are separated cleanly.
  • Positive and Negative prompts give high‑level camera and quality guidance. Keep them descriptive and factual to maintain identity and lighting continuity.
  • Segment length and overlap control how many frames are generated per pass and how smoothly segments blend. Longer segments reduce the number of passes, and more overlap softens transitions.
  • Queue the canvas to generate and save the dialogue video. Two Save Video nodes use different filename prefixes so dialogue and monologue renders do not overwrite each other.

Single‑Speaker Monologue · bypassed by default

  • Load a single‑person reference image and one audio file. The same voice track drives duration, conditioning, and the final soundtrack.
  • The single‑speaker LongCat node uses the whole image, so no masking is required. Keep the subject front‑facing with moderate headroom for best lip readability.
  • Adjust Positive and Negative prompts to match your on‑camera style. Then queue the canvas to render and save the monologue video.

Switching between modes

  • In the Groups menu, run only one group at a time. Set the mode you want to Always and set the other to Bypass. When returning to Dialogue, keep the Optional OPSD blend bypassed unless you specifically want that variant.

Key nodes in Comfyui LongCat Video Avatar 1.6 ComfyUI Workflow#

LongCatAvatarToVideo (#100) · Dialogue

  • The core generator for two‑speaker scenes. It takes the start image, two encoded audio tracks, and two SAM masks so each voice animates the correct face. Tune only what matters: segment length versus overlap for temporal smoothness, and audio strength if you need crisper mouth articulation. The node also receives a frame offset from the loop so each segment lands at the correct position in time. See the native node introduction for background. ComfyUI PR

LongCatAvatarToVideo (#300) · Monologue

  • The single‑speaker variant drives one face from one audio track. Use it when only one subject is visible. The same guidance applies: adjust segment length and overlap for your script length, and keep audio strength moderate to balance lip accuracy and facial stability. ComfyUI PR

SAM3_Detect (#147) Speaker A mask and SAM3_Detect (#150) Speaker B mask

  • These detect people in the start image and yield clean masks. They accept brief text labels, which you can change to reflect your subjects. If the masks bleed, nudge the prompts or provide simple boxes or points; use the built‑in mask previews to verify separation. For SAM background, see the original paper and repo. Paper · Repo

StartLoop · Segments (subgraph “Loop Start”)

  • Automates long scripts by computing how many segments are needed from audio duration, segment length, and overlap. It increments the seed between passes for natural variation, and outputs a frame offset so each pass starts at the correct timeline position. The loop feeds previous frames back into generation to maintain continuity across cuts.

SamplerCustom pairs with BasicScheduler and SplitSigmas

  • The workflow uses a two‑stage sampler per segment, operating on higher then lower noise ranges for motion stability. Keep the default sampler and schedule unless you have a reason to experiment; the shipped configuration matches the official LongCat release behavior. Weights · Model card

Optional extras#

  • Identity first. Use a sharp, front‑facing reference with neutral lips and consistent lighting. Small head turns are fine; large yaw or occlusions reduce lip clarity.
  • Dialogue setup. If your speakers overlap or cross, adjust the SAM prompts or refine the masks so each voice controls only one face.
  • Audio prep. Use clean, mono WAV files without heavy noise reduction. Trim leading and trailing silences to keep timing tight.
  • Prompting. Keep Positive prompts concrete and scene‑level. Use the Negative prompt to rule out artifacts like overexposure, jitter, or frozen mouth.
  • Variants. The Optional OPSD blend is provided for experimentation but is bypassed by default to mirror the published pipeline. Weights

Acknowledgements#

This workflow implements and builds upon the following works and resources. We gratefully acknowledge the Meituan LongCat team for the LongCat-Video-Avatar 1.6 model, Comfy-Org for the LongCat-Video-Avatar weights for ComfyUI, and Comfy-Org/ComfyUI for the native LongCat Avatar nodes and source workflow for their contributions and maintenance. For authoritative details, please refer to the original documentation and repositories linked below.

Resources#

Note: Use of the referenced models, datasets, and code is subject to the respective licenses and terms provided by their authors and maintainers.

RunComfy
Copyright 2026 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.