HuMo: Human-Centric Video Generation with Audio Conditioning

HuMo by Phantom-video is a human-centric video generation model built on Wan2.1. Uses whisper audio conditioning for human video generation with 14B and 1.7B sizes.

H

HuMo

Wan2.1Human-CentricAudio ConditioningWhisper

Human-centric video generation model built on Wan2.1 with whisper audio conditioning. Supports text, image, and audio inputs for fine-grained controllable human video generation. Available in 14B fp16 and 1.7B fp16 sizes.

DeveloperPhantom-video
ArchitectureWan2.1-based with multi-modal conditioning
Model Sizes14B fp16, 1.7B fp16
CapabilitiesHuman-centric video generation, text/image/audio conditioning
Audio ConditionerWhisper (via westlast/whisper)

Overview

HuMo (Human Motion) by Phantom-video is a unified human-centric video generation framework built on Wan2.1. It produces high-quality, fine-grained, and controllable human videos from multimodal inputs including text, images, and audio. The whisper audio conditioning enables precise synchronization of human motion with audio cues, making it suitable for talking head generation, gesture synthesis, and expressive human animation.

The model is available in two sizes: the full 14B fp16 variant for maximum quality and the lightweight 1.7B fp16 variant for consumer GPU compatibility.

ComfyUI Integration

HuMo is compatible with ComfyUI workflows. The whisper audio model for conditioning can be found at westlast/whisper on Hugging Face.

Guides and workflows related to this model series.

No articles found.