# bytedance/USO

[CVPR 2026] 🔥🔥 Official Repo of USO: Unified Style and Subject-Driven Generation via Disentangled and Reward Learning

Repository: https://github.com/bytedance/USO
Canonical: https://ross.abutalabs.com/products/uso
Homepage: https://bytedance.github.io/USO/
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2025-09-12T08:11:32+00:00

## Health v2 (maintenance only)
Score: 36/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 41, release rhythm 35, longevity 26
- inputs: {"age_days": 376, "days_push": 355, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1235, forks 77 (observed 2026-08-28T04:04:04.847447+00:00)

## What it is
USO is ByteDance's open-source unified style- and subject-driven image generation model based on diffusion (FLUX), combining any subject with any style via disentangled and reward learning. It ships inference code, model weights, a Hugging Face demo, ComfyUI workflows, and plans for training code and datasets.

## Use cases
- generate images of a specific subject in a specific artistic style
- style transfer that preserves subject identity
- create stylized portraits with natural, non-plastic faces
- combine reference character with reference art style
- run subject-driven customization on a consumer GPU
- use USO in ComfyUI workflows

## When to choose
- you need both subject consistency and style fidelity in one model
- you want a research-grade open model with weights and benchmarks (USO-Bench)
- you work with ComfyUI or Hugging Face and want drop-in FLUX-based customization

## When to avoid
- you need text-to-image generation without reference images
- you cannot run GPU inference (~16GB VRAM minimum in fp8 mode)
- you need training code today - it is promised but not yet fully released

## Facets
- artifact type: library
- maturity: active
- function: image-processing, machine-learning, stable-diffusion
- domain: image-processing, artificial-intelligence, computer-vision
- platform: python
- tags: style-transfer, subject-driven-generation, diffusion-model, flux, comfyui, reward-learning, cvpr-paper, customization, gpu, docker

## Member repositories
- bytedance/USO (main) score 36

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:04.847447+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T08:22:04.163381+00:00, confidence not recorded.
  - readme: https://github.com/bytedance/USO (fetched 2026-08-28T04:04:04.847447+00:00, sha e14f7c91acef)
  - homepage: https://bytedance.github.io/USO/ (fetched 2026-08-29T12:21:40.994218+00:00, sha 502a5eb35454)
- Data as of 2026-08-30T08:39:29.467469+00:00.