# ByteDance Releases HuMo for Human Video Generation

> The new open-source model specializes in creating realistic videos of people, separating appearance from motion for greater control.

Published by The Open Weights on Sep 10, 2025. Canonical: https://theopenweights.com/news/humo-f7h7

## Key facts

- Company: ByteDance
- Model: HuMo
- Version: HuMo
- Category: Image → Video
- Modalities: Image → Video
- License: Apache 2.0 (Open weights, commercial use allowed)
- Resolution: 480P & 720P
- Significance: notable
- Published: 2025-09-10
- Last verified: 2026-06-17
- Hugging Face: https://huggingface.co/bytedance-research/HuMo
- Canonical URL: https://theopenweights.com/news/humo-f7h7

ByteDance Research has released HuMo, an open-source model focused on a notoriously difficult task in AI: generating realistic videos of humans. The model can create short video clips from either a text description or a reference image, marking a new entry in the competitive field of AI video synthesis. [The weights and code are available on Hugging Face](https://huggingface.co/bytedance-research/HuMo) under a permissive Apache 2.0 license.

## A Focus on Motion and Anatomy

Unlike general-purpose video models, HuMo is specifically designed to understand and render the human form in motion. According to the project's documentation, it uses a diffusion-based architecture that separates a subject's appearance from their movement. A "human prior encoder" helps maintain anatomical consistency, while a "motion-guidance module" allows for more precise control over the action in the generated clip.

This specialized approach enables several key capabilities:

*   **Text-to-Video:** Generating a video of a person performing an action described in a prompt.
*   **Image-to-Video:** Animating a person from a single still photograph.
*   **Motion Control:** Guiding the generation process with specific motion sequences for more directed outputs.

The release is significant because generating plausible human movement without uncanny or distorted results remains a major hurdle for AI video. By focusing on this specific domain and releasing the model openly, ByteDance provides researchers and creators with a powerful new tool. The Apache 2.0 license further encourages experimentation and potential integration into commercial applications, from creative software to virtual character animation.

## Get the model

- [Hugging Face](https://huggingface.co/bytedance-research/HuMo)

## Sources

- [bytedance-research/HuMo](https://huggingface.co/bytedance-research/HuMo) — Hugging Face, Sep 10, 2025

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "ByteDance Releases HuMo for Human Video Generation", The Open Weights, Sep 10, 2025, https://theopenweights.com/news/humo-f7h7