# 😮 Highlights
### 1. All Resources Fully Open-Sourced
- We fully open-source the models, data, training and evaluation code to facilitate rapid community exploration of unified architectures.
- We curate 10+ CV downstream tasks, including canny, depth, sketch, MLSD, segmentation and so on.
- We annotate 286K long-caption samples using [Qwen2-VL-72B](https://huggingface.co/Qwen/Qwen2-VL-72B-Instruct). We use GPT-4o to filter [ImgEdit](https://github.com/PKU-YuanGroup/ImgEdit), result in 724K high-quality editing samples (all shortedge ≥ 1024 pix). Additionally, we organize and filter existing open-sourced datasets. The details can be found [here](https://github.com/PKU-YuanGroup/UniWorld/tree/main?tab=readme-ov-file#data-details).
### 2. Contrastive Semantic Encoders as Reference Control Signals
- Unlike prior approaches that use VAE-encoded reference images for low-level control, we advocate using contrastive visual encoders as control signals for reference images.
- For such encoders, we observe that as resolution increases, global features approach saturation and model capacity shifts toward preserving fine details, which is crucial for maintaining fidelity in non-edited regions.
### 3. Image Priors via VLM Encoding Without Learnable Tokens
- We find that multimodal features encoded by VLMs can interpret instructions while retaining image priors. Due to causal attention, the format `
# 🤗 Demo
### Gradio Web UI
Highly recommend trying out our web demo by the following command.
```bash
MODEL_PATH="path/to/model"
FLUX_PATH="path/to/flux"
SIGLIP_PATH="path/to/siglip"
CUDA_VISIBLE_DEVICES=0 python -m univa.serve.gradio_web_server \
--model_path ${MODEL_PATH} \
--flux_path ${FLUX_PATH} \
--siglip_path ${SIGLIP_PATH}
```
### CLI Inference
```bash
MODEL_PATH="path/to/model"
FLUX_PATH="path/to/flux"
SIGLIP_PATH="path/to/siglip"
CUDA_VISIBLE_DEVICES=1 python -m univa.serve.cli \
--model_path ${MODEL_PATH} \
--flux_path ${FLUX_PATH} \
--siglip_path ${SIGLIP_PATH}
```
### ComfyUI
Coming soon...
# ⚙️ Requirements and Installation
1. Clone this repository and navigate to UniWorld folder
```
git clone https://github.com/PKU-YuanGroup/UniWorld
cd UniWorld
```
2. Install required packages
```
conda create -n univa python=3.10 -y
conda activate univa
pip install -r requirements.txt
```
# 🗝️ Training
### Data preparation
Download the data from [LanguageBind/UniWorld-V1](https://huggingface.co/datasets/LanguageBind/UniWorld-V1). The dataset consists of two parts: source images and annotation JSON files.
Prepare a `data.txt` file in the following format:
1. The first column is the root path to the image.
2. The second column is the corresponding annotation JSON file.
3. The third column indicates whether to enable the region-weighting strategy. We recommend setting it to True for edited data and False for others.
```
data/BLIP3o-60k,json/blip3o_t2i_58859.json,false
data/coco2017_caption_canny-236k,coco2017_canny_236574.json,false
data/imgedit,json/imgedit/laion_add_part0_edit.json,true
```
We provide a simple online verification tool to check whether your paths are set in `data.txt` correctly.
```
python univa/serve/check_data.py
```
### Data details
- [BLIP3o-60k](https://huggingface.co/datasets/BLIP3o/BLIP3o-60k): We add text-to-image instructions to half of the data. [108 GB storage usage.]
- [OSP1024-286k](https://huggingface.co/datasets/LanguageBind/UniWorld-V1/tree/main/data/OSP1024-286k): Sourced from internal data of the [Open-Sora Plan](https://github.com/PKU-YuanGroup/Open-Sora-Plan), with captions generated using [Qwen2-VL-72B](https://huggingface.co/Qwen/Qwen2-VL-72B-Instruct). Images have an aspect ratio between 3:4 and 4:3, aesthetic score ≥ 6, and a short side ≥ 1024 pixels. [326 GB storage usage.]
- [imgedit-724k](https://huggingface.co/datasets/sysuyy/ImgEdit/tree/main): Data is filtered using GPT-4o, retaining approximately half. [2.1T storage usage.]
- [OmniEdit-368k](https://huggingface.co/datasets/TIGER-Lab/OmniEdit-Filtered-1.2M): For image editing data, samples with edited regions smaller than 1/100 were filtered out; images have a short side ≥ 1024 pixels. [204 GB storage usage.]
- [SEED-Data-Edit-Part1-Openimages-65k](https://huggingface.co/datasets/AILab-CVC/SEED-Data-Edit-Part1-Openimages): For image editing data, samples with edited regions smaller than 1/100 were filtered out. Images have a short side ≥ 1024 pixels. [10 GB storage usage.]
- [SEED-Data-Edit-Part2-3-12k](https://huggingface.co/datasets/AILab-CVC/SEED-Data-Edit-Part2-3): For image editing data, samples with edited regions smaller than 1/100 were filtered out. Images have a short side ≥ 1024 pixels. [10 GB storage usage.]
- [PromptfixData-18k](https://huggingface.co/datasets/yeates/PromptfixData): For image restoration data and some editing data, samples with edited regions smaller than 1/100 were filtered out. Images have a short side ≥ 1024 pixels. [9 GB storage usage.]
- [StyleBooth-11k](https://huggingface.co/scepter-studio/stylebooth): For transfer style data, images have a short side ≥ 1024 pixels. [4 GB storage usage.]
- [Ghibli-36k](https://huggingface.co/datasets/LanguageBind/UniWorld-V1/tree/main/data/Ghibli-36k): For transfer style data, images have a short side ≥ 1024 pixels. **Warning: This data has not been quality filtered.** [170 GB storage usage.]
- [viton_hd-23k](https://huggingface.co/datasets/forgeml/viton_hd): Converted from the source data into an instruction dataset for product extraction. [1 GB storage usage.]
- [deepfashion-27k](https://huggingface.co/datasets/lirus18/deepfashion): Converted from the source data into an instruction dataset for product extraction. [1 GB storage usage.]
- [shop_product-23k](https://huggingface.co/datasets/LanguageBind/UniWorld-V1/tree/main/data/shop_product-23k): Sourced from internal data of the [Open-Sora Plan](https://github.com/PKU-YuanGroup/Open-Sora-Plan), focusing on product extraction and virtual try-on, with images having a short side ≥ 1024 pixels. [12 GB storage usage.]
- [coco2017_caption_canny-236k](https://huggingface.co/datasets/gebinhui/coco2017_caption_canny): img->canny & canny->img [25 GB storage usage.]
- [coco2017_caption_depth-236k](https://huggingface.co/datasets/gebinhui/coco2017_caption_depth): img->depth & depth->img [8 GB storage usage.]
- [coco2017_caption_hed-236k](https://huggingface.co/datasets/gebinhui/coco2017_caption_hed): img->hed & hed->img [13 GB storage usage.]
- [coco2017_caption_mlsd-236k](https://huggingface.co/datasets/gebinhui/coco2017_caption_mlsd): img->mlsd & mlsd->img [ GB storage usage.]
- [coco2017_caption_normal-236k](https://huggingface.co/datasets/gebinhui/coco2017_caption_normal): img->normal & normal->img [10 GB storage usage.]
- [coco2017_caption_openpose-62k](https://huggingface.co/datasets/wangherr/coco2017_caption_openpose): img->pose & pose->img [2 GB storage usage.]
- [coco2017_caption_sketch-236k](https://huggingface.co/datasets/wangherr/coco2017_caption_sketch): img->sketch & sketch->img [15 GB storage usage.]
- [unsplash_canny-20k](https://huggingface.co/datasets/wtcherr/unsplash_10k_canny): img->canny & canny->img [2 GB storage usage.]
- [open_pose-40k](https://huggingface.co/datasets/raulc0399/open_pose_controlnet): img->pose & pose->img [4 GB storage usage.]
- [mscoco-controlnet-canny-less-colors-236k](https://huggingface.co/datasets/hazal-karakus/mscoco-controlnet-canny-less-colors): img->canny & canny->img [13 GB storage usage.]
- [coco2017_seg_box-448k](https://huggingface.co/datasets/LanguageBind/UniWorld-V1/tree/main/data/coco2017_seg_box-448k): img->detection & img->segmentation (mask), instances with regions smaller than 1/100 were filtered out. We visualise masks on the original image as gt-image. [39 GB storage usage.]
- [viton_hd-11k](https://huggingface.co/datasets/forgeml/viton_hd): img->pose [1 GB storage usage.]
- [deepfashion-13k](https://huggingface.co/datasets/lirus18/deepfashion): img->pose [1 GB storage usage.]
```
cd univa/eval/geneval
# follow the instruction in univa/eval/geneval/README.md
```
```
cd univa/eval/wise
# follow the instruction in univa/eval/wise/README.md
```
```
cd univa/eval/genai
# follow the instruction in univa/eval/genai/README.md
```
```
cd univa/eval/dpgbench
# follow the instruction in univa/eval/dpgbench/README.md
```
```
cd univa/eval/imgedit
# follow the instruction in univa/eval/imgedit/README.md
```
```
cd univa/eval/gdit
# follow the instruction in univa/eval/gdit/README.md
```
# 💡 How to Contribute
We greatly appreciate your contributions to the UniWorld open-source community and helping us make it even better than it is now!
For more details, please refer to the [Contribution Guidelines](docs/Contribution_Guidelines.md).
# 👍 Acknowledgement and Related Work
* [ImgEdit](https://github.com/PKU-YuanGroup/ImgEdit): ImgEdit is a large-scale, high-quality image-editing dataset comprising 1.2 million carefully curated edit pairs.
* [Open-Sora Plan](https://github.com/PKU-YuanGroup/Open-Sora-Plan): An open‑source text-to-image/video foundation model, which provides a lot of caption data.
* [SEED-Data-Edit](https://huggingface.co/datasets/AILab-CVC/SEED-Data-Edit): A hybrid dataset for instruction-guided image editing.
* [Qwen2.5-VL](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct): The new flagship vision-language model of Qwen.
* [FLUX.1-Redux-dev](https://huggingface.co/black-forest-labs/FLUX.1-Redux-dev): Given an input image, FLUX.1 Redux can reproduce the image with slight variation, allowing to refine a given image.
* [SigLIP 2](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md): New multilingual vision-language encoders.
* [Step1X-Edit](https://github.com/stepfun-ai/Step1X-Edit): A state-of-the-art image editing model.
* [BLIP3-o](https://github.com/JiuhaiChen/BLIP3o): A unified multimodal model that combines the reasoning and instruction following strength of autoregressive models with the generative power of diffusion models.
* [BAGEL](https://github.com/ByteDance-Seed/Bagel): An open‑source multimodal foundation model with 7B active parameters (14B total) trained on large‑scale interleaved multimodal data.
# 🔒 License
* See [LICENSE](LICENSE) for details. The FLUX weights fall under the [FLUX.1 [dev] Non-Commercial License](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md).
## ✨ Star History
[](https://star-history.com/#PKU-YuanGroup/UniWorld&Date)
# ✏️ Citing
```bibtex
@misc{lin2025uniworldhighresolutionsemanticencoders,
title={UniWorld: High-Resolution Semantic Encoders for Unified Visual Understanding and Generation},
author={Bin Lin and Zongjian Li and Xinhua Cheng and Yuwei Niu and Yang Ye and Xianyi He and Shenghai Yuan and Wangbo Yu and Shaodong Wang and Yunyang Ge and Yatian Pang and Li Yuan},
year={2025},
eprint={2506.03147},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2506.03147},
}
```
```bibtex
@article{niu2025wise,
title={Wise: A world knowledge-informed semantic evaluation for text-to-image generation},
author={Niu, Yuwei and Ning, Munan and Zheng, Mengren and Lin, Bin and Jin, Peng and Liao, Jiaqi and Ning, Kunpeng and Zhu, Bin and Yuan, Li},
journal={arXiv preprint arXiv:2503.07265},
year={2025}
}
```
```bibtex
@article{lin2024open,
title={Open-Sora Plan: Open-Source Large Video Generation Model},
author={Lin, Bin and Ge, Yunyang and Cheng, Xinhua and Li, Zongjian and Zhu, Bin and Wang, Shaodong and He, Xianyi and Ye, Yang and Yuan, Shenghai and Chen, Liuhan and others},
journal={arXiv preprint arXiv:2412.00131},
year={2024}
}
```
# 🤝 Community contributors
Text-to-Image Generation
Image Editing
Extract & Try-on
### Training
#### Prepare pretrained weights
Download [black-forest-labs/FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev) to `$FLUX_PATH`.
Download [Qwen/Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct) to `$QWENVL_PATH`. We also support other sizes of Qwen2.5-VL.
```
SAVE_PATH="path/to/save/UniWorld-Qwen2.5-VL-7B-Instruct-FLUX.1-dev-fp32"
python scripts/make_univa_qwen2p5vl_weight.py \
--origin_flux_ckpt_path $FLUX_PATH \
--origin_qwenvl_ckpt_path $QWENVL_PATH \
--save_path ${SAVE_PATH}
```
```
# stage1
bash scripts/denoiser/flux_qwen2p5vl_7b_vlm_stage1_512.sh
```
Download [flux-redux-siglipv2-512.bin](https://huggingface.co/LanguageBind/UniWorld-V1/resolve/main/flux-redux-siglipv2-512.bin?download=true) and set its path to `pretrained_siglip_mlp_path` in `stage2.yaml`. The weight is sourced from [ostris/Flex.1-alpha-Redux](https://huggingface.co/ostris/Flex.1-alpha-Redux), we just re-organize the weight.
```
# stage2
bash scripts/denoiser/flux_qwen2p5vl_7b_vlm_stage2_512.sh
```
# ⚡️ Evaluation
### Text-to-Image Generation
Image Perception
GenEval
WISE
GenAI-Bench
### Image Editing
DPG-Bench
ImgEdit
# 📊 Benchmarks
GEdit
This model is presented in the paper: [UniWorld: High-Resolution Semantic Encoders for Unified Visual Understanding and Generation](https://huggingface.co/papers/2506.03147)