| --- |
| license: apache-2.0 |
| language: |
| - en |
| base_model: |
| - Qwen/Qwen-Image-Edit-2509 |
| - Qwen/Qwen-Image-Edit-2511 |
| - sensenova/SenseNova-U1-8B-MoT |
| pipeline_tag: any-to-any |
| tags: |
| - style transfer |
| - stylization |
| --- |
| <div align="center"> |
|
|
| <h1><b>TeleStyle V2: Beyond Content-Preserving Style Transfer with Self-Distillation and Distribution-Matching-Distillation</b></h1> |
|
|
| </div> |
|
|
| <div align="center"> |
| Shiwen Zhang, Yifan Xu, Haibin Huang, Chi Zhang, Xuelong Li |
| <p> |
| Institute of Artificial Intelligence, China Telecom (TeleAI) |
| </p> |
| </div> |
| <div align="center"> |
| [<a href="https://witcherofresearch.github.io/TeleStyleV2/" target="_blank">Project Page</a>] |
| [<a href="https://arxiv.org/abs/2606.20709" target="_blank">Paper</a>] |
| [<a href="https://huggingface.co/Tele-AI/TeleStyleV2" target="_blank">Model</a>] |
| [<a href="https://github.com/Tele-AI/TeleStyleV2" target="_blank">Code</a>] |
| [<a href="https://huggingface.co/spaces/witcherderivia/TeleStyleV2" target="_blank">Demo</a>] |
| </div> |
| |
|
|
| ## 🔔News |
| - [2026-07] Released TeleStyleV2-SenseNova, [Code](https://huggingface.co/spaces/witcherderivia/TeleStyle-SenseNova/tree/main), [Model](https://huggingface.co/Tele-AI/TeleStyleV2), [Demo](https://huggingface.co/spaces/witcherderivia/TeleStyle-SenseNova), reinforces SenseNova U1 for Content-Preserving Style Transfer and preserves its general image editing capability. This experimental model is trained in pixel space, making the SFT quite hard and still having much space to improve. The model supports 1MP to 4MP. This is maybe the first unofficial SFT model of SenseNovaU1. |
| - June, 2026: We found that TeleStyleV2 has been integrated by fal.ai at [fal-ai/telestyle-v2](https://fal.ai/models/fal-ai/telestyle-v2/api) and mlx-community at [mlx-community/TeleStyleV2-Qwen-Image-Edit-2511-bf16](https://huggingface.co/mlx-community/TeleStyleV2-Qwen-Image-Edit-2511-bf16). Thanks to the community for supporting the project. |
| - June 25, 2026: We release [TeleStyleV2-QIE-2511-Lora demo](https://huggingface.co/spaces/witcherderivia/TeleStyleV2-QIE2511). |
| - June 11, 2026: We release a [free online demo for TeleStyleV2-QIE-2509-Lora-DMD ](https://huggingface.co/spaces/witcherderivia/TeleStyleV2) for diffusers model. We also release the diffsynth version of [TeleStyleV2-QIE-2509-Lora](https://huggingface.co/Tele-AI/TeleStyleV2/blob/main/diffsynth-TeleStyleV2-QIE-2509-Lora-bf16.safetensors). Please light a star to support this project if you find the demo useful. |
| - June 10, 2026: We release the <a href="http://arxiv.org/abs/2601.20175" target="_blank">technical report and project page </a> of TeleStyleV2. We release the diffusers version of [TeleStyleV2-QIE-2509-Lora](https://huggingface.co/Tele-AI/TeleStyleV2/blob/main/diffusers-TeleStyleV2-QIE-2509-Lora-bf16.safetensors). |
| - [2026-06]: Released [Code](https://github.com/Tele-AI/TeleStyleV2), [Model](https://huggingface.co/Tele-AI/TeleStyleV2), [Demo] (https://huggingface.co/spaces/witcherderivia/TeleStyleV2). |
|
|
| ## How to use |
|
|
| - Please refer to [**🔗 GitHub**](https://github.com/Tele-AI/TeleStyleV2) for usage. |
|
|
|
|
| ## 🌟 Citation |
|
|
| If you find TeleStyleV2 useful for your research and applications, please cite using this BibTeX: |
|
|
| ```bibtex |
| @article{telestylev2, |
| title={TeleStyle V2: Beyond Content-Preserving Style Transfer with Self-Distillation and Distribution-Matching-Distillation}, |
| author={Shiwen Zhang, Yifan Xu, Haibin Huang, Chi Zhang, Xuelong Li}, |
| journal={TeleAI}, |
| year={2026}, |
| } |
| ``` |
