---
title: Fill2SR
canonical_url: "https://www.modelscope.ai/models/yixingfu/Fill2SR"
md_url: "https://www.modelscope.ai/models/yixingfu/Fill2SR.md"
repository: yixingfu/Fill2SR
last_updated: 2026-10-11
license: other
pipeline_tag: image-to-image
tasks:
  - image-to-image
base_model:
  - black-forest-labs/FLUX.1-Fill-dev
  - black-forest-labs/FLUX.1-Redux-dev
base_model_relation: adapter
parameters: 717.2M
tensor_type:
  - BF16
library_name:
  - lora
  - safetensors
language:
  - en
downloads: 4
stars: 0
tags:
  - super-resolution
  - image-restoration
  - flux
  - lora
  - eccv-2026
---

# Fill2SR

> Fill2SR - An open-source model by yixingfu on ModelScope. Official Ours-Full checkpoint for Fill2SR (ECCV 2026).

yixingfu/Fill2SR is a 717.2M-parameter image-to-image model on ModelScope. licensed under other. derived from black-forest-labs/FLUX.1-Fill-dev、black-forest-labs/FLUX.1-Redux-dev.

- **Repository**: yixingfu/Fill2SR
- **License**: other
- **Tasks**: image-to-image
- **Parameters**: 717.2M
- **Base model**: black-forest-labs/FLUX.1-Fill-dev, black-forest-labs/FLUX.1-Redux-dev
- **Tags**: super-resolution, image-restoration, flux, lora, eccv-2026
- **Downloads**: 4
- **Stars**: 0
- **Last updated**: 2026-10-11

Source: https://www.modelscope.ai/models/yixingfu/Fill2SR

---

# Fill2SR — Ours-Full

This repository hosts the **Ours-Full** LoRA from [Fill2SR: Repurposing Inpainting Diffusion Transformers for Real-World Super-Resolution](https://doi.org/10.1007/978-3-032-37556-8_24), ECCV 2026, by Xingfu Yi and Xiaoxue Yu. It is the paper's full model with RCDT, built on the IIEA super-resolution interface. The weights are intended for research on real-world image super-resolution. [Inference code and instructions](https://github.com/Xingfu-Yi/Fill2SR) are in the official GitHub repository.

The released checkpoint is `sr-full.safetensors`, the paper's **Ours-Full** LoRA. Base FLUX checkpoints are downloaded separately. Training code and Ours-Base weights are not included.

## Resources

- [Paper](https://doi.org/10.1007/978-3-032-37556-8_24)
- [Code and inference instructions](https://github.com/Xingfu-Yi/Fill2SR)
- Model repositories: [Hugging Face](https://huggingface.co/yixingfu/Fill2SR), [ModelScope China](https://modelscope.cn/models/yixingfu/Fill2SR), [ModelScope International](https://modelscope.ai/models/yixingfu/Fill2SR)

Each repository distributes the same Ours-Full checkpoint. Access requirements are managed by the respective hosting platform.

## Dependencies and usage

Download the two base models from the official Black Forest Labs repositories after satisfying their access requirements:

| Base model | Hugging Face | ModelScope China |
| --- | --- | --- |
| FLUX.1-Fill-dev | [Download](https://huggingface.co/black-forest-labs/FLUX.1-Fill-dev) | [Download](https://modelscope.cn/models/black-forest-labs/FLUX.1-Fill-dev) |
| FLUX.1-Redux-dev | [Download](https://huggingface.co/black-forest-labs/FLUX.1-Redux-dev) | [Download](https://modelscope.cn/models/black-forest-labs/FLUX.1-Redux-dev) |

Place these models and the Fill2SR LoRA under `models/` as described in the [GitHub README](https://github.com/Xingfu-Yi/Fill2SR#pretrained-models):

```text
models/
├── FLUX.1-Fill-dev/
├── FLUX.1-Redux-dev/
└── sr-full.safetensors
```

The reference example uses RealLQ250 `113.png`, the **2K preset, 28 steps and seed 2026**. See the [inference instructions](https://github.com/Xingfu-Yi/Fill2SR#inference) for running the model.


Checkpoint integrity is recorded in [checkpoint.json](checkpoint.json). SHA-256 for `sr-full.safetensors`:

```text
713a7c8e6cf6ca4265b0d699ecaff1dc0d13039b6ac171431dc68c3af695fbe6
```

The LoRA is intended for image restoration. Generated details may differ from the input and should be checked before use in contexts where fidelity matters. Results depend on the input degradation and inference settings.

## License and attribution

This LoRA was made by modifying FLUX.1-Fill-dev using LoRA fine-tuning. It is distributed under the [FLUX.1 [dev] Non-Commercial License](LICENSE.md) from Black Forest Labs Inc. The license applies to this derivative and limits use to non-commercial and non-production purposes. Recipients receive their rights to the underlying FLUX.1 [dev] models and derivatives directly from Black Forest Labs under that license. Read the full license, including all restrictions and disclaimers, before downloading or using the weights.

> The FLUX.1 [dev] Model is licensed by Black Forest Labs Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs Inc.
> IN NO EVENT SHALL BLACK FOREST LABS INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

The applicable FLUX.1 [dev] Model has been modified by Xingfu Yi and Xiaoxue Yu to create the Fill2SR Ours-Full LoRA. Fill2SR is an independent research release; it is not an official Black Forest Labs product and is not endorsed, approved or validated by Black Forest Labs. This repository contains the required copy of the upstream license. The upstream license applies to the checkpoint on every hosting platform and to downloaded copies.

## Citation

If you find Fill2SR useful in your research, please cite:

```bibtex
@inproceedings{yi2026fill2sr,
  author={Yi, Xingfu and Yu, Xiaoxue},
  title={{Fill2SR}: Repurposing Inpainting Diffusion Transformers for Real-World Super-Resolution},
  booktitle={Computer Vision -- ECCV 2026},
  year={2026},
  doi={10.1007/978-3-032-37556-8_24}
}
```

Contact: yixingfu.research@gmail.com.