# HuMo
HuMo is an open-source, human-centric video generation framework from ByteDance Research that produces controllable videos of people from combined text, image and audio inputs, built on the Wan 2.1 text-to-video model and released in September 2025.[1](https://arxiv.org/html/2509.08519v1) It is a unified human-centric video generation framework that combines subject preservation (keeping a reference person's identity consistent) with synchronized audio-driven motion.[2](https://huggingface.co/bytedance-research/HuMo) The maker, its founders, and ByteDance's consumer video products are covered in separate articles.
| Key fact | Detail |
| --- | --- |
| Developer | ByteDance Research[3](https://comfyui-wiki.com/en/models/humo/humo) |
| First release | Technical report and project page September 9, 2025; 17B weights September 10; 1.7B weights September 16, 2025[4](https://github.com/Phantom-video/HuMo) |
| Base model | Wan-2.1-1.3B and Wan-2.1-14B diffusion transformers; Whisper-large-v3 audio encoder[1](https://arxiv.org/html/2509.08519v1) • [2](https://huggingface.co/bytedance-research/HuMo) |
| Sizes | 1.7B and 17B, in FP16 and FP8[3](https://comfyui-wiki.com/en/models/humo/humo) |
| Output | 480P (both sizes) and 720P (17B); trained on 97-frame clips at 25 fps, 480×832[2](https://huggingface.co/bytedance-research/HuMo) • [1](https://arxiv.org/html/2509.08519v1) |
| License | Apache 2.0[4](https://github.com/Phantom-video/HuMo) |
| Hardware | 1.7B runs on a 32 GB GPU, generating a 480P video in about 8 minutes; 17B supports multi-GPU inference via FSDP[4](https://github.com/Phantom-video/HuMo) • [2](https://huggingface.co/bytedance-research/HuMo) |
| Peer review | Published in the AAAI-40 proceedings[5](https://doi.org/10.1609/aaai.v40i4.37285) |
## What HuMo is
HuMo is a conditional video generator for people: it takes a text prompt, an optional reference image of a person, and an optional audio track, and produces a video in which the described person appears, keeps the reference identity, and moves in sync with the audio.[2](https://huggingface.co/bytedance-research/HuMo) It supports three conditioning modes: text-image, text-audio (which removes the need for a reference image), and text-image-audio.[4](https://github.com/Phantom-video/HuMo)
HuMo generates the video itself from a still image or from text plus audio, while also preserving subject identity across frames, which is why its evaluation covers both image quality and identity metrics and audio-sync metrics.[1](https://arxiv.org/html/2509.08519v1)
## Release timeline and versions
The project page and technical report appeared on arXiv on September 9, 2025. ByteDance released the HuMo-17B weights and inference code on September 10, 2025, and the lightweight HuMo-1.7B weights on September 16, 2025.[4](https://github.com/Phantom-video/HuMo) ComfyUI-Wan merged support for the 17B model on September 13, 2025, and ComfyUI officially supported HuMo-1.7B on September 17, 2025, with the 17B model runnable on an NVIDIA 3090.[4](https://github.com/Phantom-video/HuMo)
The team followed the models with training data: the Stage-1 subject-preservation dataset on September 30, 2025, and on December 23, 2025 the HuMoSet dataset of 670,000 video samples with diverse reference images, dense video captions, and strict audio-visual synchronization.[4](https://github.com/Phantom-video/HuMo) A longer-generation checkpoint, HuMo-Longer, was announced for October 2025; the record does not confirm whether it shipped.[4](https://github.com/Phantom-video/HuMo) The paper itself was peer-reviewed and published in the AAAI-40 proceedings (volume 40, issue 4).[5](https://doi.org/10.1609/aaai.v40i4.37285)
One date discrepancy exists in the record: ComfyUI Wiki lists an August 2025 release date, while the official repository dates the technical report September 9, 2025 and the 17B weights September 10, 2025; the official dates are used here.[3](https://comfyui-wiki.com/en/models/humo/humo) • [4](https://github.com/Phantom-video/HuMo)
## Architecture and training
HuMo builds on the Wan-2.1 diffusion transformer (DiT) text-to-video backbones, in 1.3B and 14B parameter versions, which use a 3D variational autoencoder (VAE) to compress video into a compact latent space and inject text via cross-attention; training uses flow matching, and all video is resampled to 25 fps at 480×832 resolution.[1](https://arxiv.org/html/2509.08519v1) • [5](https://doi.org/10.1609/aaai.v40i4.37285) Audio is encoded with Whisper-large-v3, with an optional audio separator, and the 17B model supports multi-GPU inference through FSDP plus sequence parallelism.[2](https://huggingface.co/bytedance-research/HuMo)
The reference image is added through a minimal-invasive image injection strategy that avoids structural changes to the DiT backbone and confines parameter updates to a limited subset of layers.[5](https://doi.org/10.1609/aaai.v40i4.37285) Audio enters through cross-attention layers with a focus-by-predicting strategy.[5](https://doi.org/10.1609/aaai.v40i4.37285)
Training follows a two-stage curriculum of 40,000 steps each. Stage 1 disables all audio-related modules and trains subject preservation only. Stage 2 enables audio cross-attention and starts with a task weighting of 80 percent subject preservation and 20 percent audio-visual sync, rising to 50/50, updating only self-attention layers and audio-related modules.[1](https://arxiv.org/html/2509.08519v1) A stage-adaptive Classifier-Free Guidance strategy dynamically adjusts guidance weights across denoising steps.[5](https://doi.org/10.1609/aaai.v40i4.37285)
## Benchmarks: vendor-reported only
All quantitative results in the record come from ByteDance's own paper. The authors evaluated against subject-to-video baselines MAGREF, HunyuanCustom, Phantom-Wan-14B, and the closed-source Kling 1.6, and against audio-visual baselines Hallo3, FantasyTalking, HunyuanCustom, and the closed-source OmniHuman-1.[1](https://arxiv.org/html/2509.08519v1) No independent evaluation of HuMo appears in the record, so the numbers below should be read as vendor claims.
On the authors' in-house 100-case benchmark, HuMo-17B scored AES 0.657, IQA 0.717, HSP 3.906, TVA 3.939, ID-Cur 0.731, ID-Glink 0.757, CLIP-I 0.687, and DINO-I 0.447, against Kling 1.6's AES 0.645, IQA 0.714, TVA 2.564, and ID-Cur 0.470.[1](https://arxiv.org/html/2509.08519v1)
On the MoCha benchmark for audio-visual synced methods, the paper reports HuMo-17B at Sync-C 6.252 and Sync-D 8.577, versus OmniHuman-1 at 6.526 and 7.784 and Hallo3 at 5.189 and 9.212. ByteDance's own table therefore shows closed-source OmniHuman-1 retaining the better Sync-C lip-sync score, with HuMo leading on Sync-D; no source in the record compares HuMo with LatentSync, Sonic, or EMO.[1](https://arxiv.org/html/2509.08519v1)
## Licensing, availability and hardware
The repository is licensed under Apache 2.0 and had 1,282 stars and 246 forks as of September 2026. Code and weights are hosted on GitHub and [Hugging Face](https://www.edgepedia.org/hugging-face), with the 1.7B and 17B models distributed in FP16 and FP8.[4](https://github.com/Phantom-video/HuMo) • [3](https://comfyui-wiki.com/en/models/humo/humo)
Hardware requirements are modest by video-generation standards. According to the repository, the 1.7B model generates a 480P video in 8 minutes on a 32 GB GPU, with lower visual quality than the 17B but nearly unaffected audio-visual sync; the 17B model runs on an NVIDIA 3090 through ComfyUI and scales to multi-GPU setups via FSDP.[4](https://github.com/Phantom-video/HuMo) • [2](https://huggingface.co/bytedance-research/HuMo) The ComfyUI integration within a week of the weight releases is the main documented adoption signal; the record contains no data on use by creators, dubbing pipelines, or avatar products beyond that.[4](https://github.com/Phantom-video/HuMo)
## References
1. HuMo: Human-Centric Video Generation via Collaborative Multi-Modal Conditioning, arXiv, https://arxiv.org/html/2509.08519v1
2. bytedance-research/HuMo, Hugging Face model card, https://huggingface.co/bytedance-research/HuMo
3. HuMo: Unified Human-Centric Video Generation by ByteDance Research, ComfyUI Wiki, https://comfyui-wiki.com/en/models/humo/humo
4. Phantom-video/HuMo, GitHub repository, https://github.com/Phantom-video/HuMo
5. HuMo: Human-Centric Video Generation via Collaborative Multi-Modal Conditioning, AAAI-40 proceedings, https://doi.org/10.1609/aaai.v40i4.37285
---
*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Modern AI: foundation models, generative AI, and the AI industry › Model families and named models › Video generation models*
*Initially written Sep 17, 2026 · Edited: Oct 11, 2026; Oct 15, 2026*
*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*
License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license