LongCat-Video-Avatar: ComfyUI Paths

LongCat-Video-Avatar: ComfyUI Paths

At a Glance

Item Value
Repository Comfy-Org/LongCat-Video-Avatar
Published 2026-11-19
License mit
Formats safetensors
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

Overview

Comfy-Org has released model files for “LongCat-Video-Avatar" compatible with ComfyUI. This model is an audio-driven video generation model that generates human videos based on audio input.

The files provided in this repository are repackaged versions of meituan-longcat/LongCat-Video-Avatar-1.6 optimized for ComfyUI. By combining multiple inputs such as text, images, and audio, users can create avatar videos with natural lip-syncing.

Specifications

The main components included in this repository are as follows:

  • Base model: meituan-longcat/LongCat-Video-Avatar-1.6
  • Supported input/output formats:
    • Audio-text-to-video – Audio-image-text-to-video – Audio-driven-video-continuation
  • Generated video specifications:
    • Supports single-character and multi-character generation. – Capable of generating long videos exceeding 10 minutes.
  • Main distributed files:
    • Diffusion model: longcat-avatar-1.6_int8_convrot.safetensors – Text Encoder: umt5_xxl_fp16.safetensors, umt5_xxl_int8_convrot.safetensors – VAE: wan_2.1_vae.safetensors – LoRA: longcat-avatar-1.6_dmd_lora_rank128_bf16.safetensors, longcat-avatar-1.6_opsd_lora_rank128_bf16.safetensors
  • License: MIT License

Performance and Quality

Regarding the performance of the original model LongCat-Video-Avatar-1.6, the model card presents results from human evaluation experiments. In the “Human Preference Arena" based on 33,051 valid votes, comparisons were made with the following six systems:

Model Name Evaluation Result
OmniHuman-2.0 1st
LongCat-Video-Avatar-1.6 2nd (Bradley–Terry Elo score: 1068.1)
LongCat-Video-Avatar 1.5 –
HeyGen –
OmniHuman-1.5 –
Kling Avatar 2.0 –

In this evaluation, LongCat-Video-Avatar-1.6 fell short of OmniHuman-2.0, but scored higher than the previous version 1.5, HeyGen, OmniHuman-1.5, and Kling Avatar 2.0.

Additionally, the model card reports the following quality improvements:

  • Improved naturalness and stability of lip animation: Reduction of overly exaggerated or unnatural mouth movements, excessively red lip coloration, and temporal instability around the mouth.
  • Stabilization of long-video generation: Suppression of error accumulation in autoregressive video continuation. This allows the generation of videos exceeding 10 minutes while maintaining visual quality, identity consistency, and temporal coherence.
  • Accelerated inference: Optimization of the end-to-end pipeline and model execution enables more efficient deployment and serving.

Strengths and Use Cases

The original model LongCat-Video-Avatar-1.6 specializes in audio-driven human (avatar) video generation. Specifically, it features the following use cases and characteristics:

  • Diverse Input Combinations: Can generate videos from audio and text only, or from a combination of audio, images, and text.
  • Video Continuation: Enables the continuation of existing videos (Video Continuation) using audio input.
  • Multi-Character Support: Supports animation generation for multiple characters as well as single characters.
  • Long Video Generation: Excels at creating long videos exceeding 10 minutes while maintaining visual quality and character consistency by suppressing error accumulation through autoregressive methods.
  • Natural Expression: Designed to enhance the naturalness of lip movements (lip-syncing) and the coordination of facial and body movements.

Distributed Files

Weight files published in Comfy-Org/LongCat-Video-Avatar, listed by this site from the Hugging Face API. Sizes are the actual file sizes.

File Size ComfyUI folder
diffusion_models/longcat-avatar-1.6_int8_convrot.safetensors 15.97GB models/diffusion_models/
loras/longcat-avatar-1.6_dmd_lora_rank128_bf16.safetensors 1.26GB models/loras/
loras/longcat-avatar-1.6_opsd_lora_rank128_bf16.safetensors 1.26GB models/loras/
text_encoders/umt5_xxl_fp16.safetensors 11.37GB models/text_encoders/
text_encoders/umt5_xxl_int8_convrot.safetensors 5.69GB models/text_encoders/
vae/wan_2.1_vae.safetensors 254MB models/vae/

The ComfyUI folder is shown only where the repository’s layout mirrors ComfyUI’s models/ directory.

How to Get It

The files in this repository are used by placing them in the respective ComfyUI model folders.

  • Distribution Format: Provided in safetensors format.
  • Placement Paths:
    • Diffusion model: ComfyUI/models/diffusion_models/ – LoRA: ComfyUI/models/loras/ – Text Encoder: ComfyUI/models/text_encoders/ – VAE: ComfyUI/models/vae/

License agreement on Hugging Face is not required to use this model.

Can You Run It Locally?

The publisher distributes this model as safetensors.

License — mit (Commercial use allowed): Permits commercial use, modification and redistribution, provided the copyright notice and license text are retained.

Compiled by this site’s code from the published formats and the license field. License summaries are not legal advice — check the publisher’s original terms before relying on them.

Related Articles

What to Read Next

Sources