---
license: other
license_name: flux-1-dev-non-commercial-license
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md
library_name: diffusers
pipeline_tag: text-to-image
base_model: black-forest-labs/FLUX.1-dev
tags:
- diffusers
- flux
- remote-sensing
- text-to-image
- sharp
- aerial-image-generation
---

# FLUX-RS

FLUX-RS is a remote sensing text-to-image checkpoint obtained by fine-tuning `black-forest-labs/FLUX.1-dev` on a curated remote sensing corpus. It is designed to improve semantic alignment and visual realism for aerial and satellite scene synthesis, especially in settings that require dense structural details such as buildings, roads, vehicles, farmland parcels, and waterfront layouts.

This model serves as the domain-specialized generative prior used in **SHARP**: *Spectrum-aware Highly-dynamic Adaptation for Resolution Promotion in Remote Sensing Synthesis*.

## Related Resources

- Research paper: [SHARP: Spectrum-aware Highly-dynamic Adaptation for Resolution Promotion in Remote Sensing Synthesis](https://arxiv.org/pdf/2603.21783)
- Official SHARP GitHub repository: [bxuanz/SHARP](https://github.com/bxuanz/SHARP)

## Model Summary

- Base model: `black-forest-labs/FLUX.1-dev`
- Model type: Diffusers `FluxPipeline`
- Domain: remote sensing text-to-image generation
- Intended use: research, evaluation, and remote sensing synthesis experiments

## Usage

### Diffusers

```python
import torch
from diffusers import FluxPipeline

pipe = FluxPipeline.from_pretrained(
    "BxuanZ/FLUX-RS",
    torch_dtype=torch.bfloat16,
)
pipe.enable_model_cpu_offload()

image = pipe(
    "Satellite imagery showing a modern downtown beside a wide river, several bridges linking both banks, office towers casting long shadows, riverside parks, and dense commercial blocks arranged along the waterfront.",
    height=1024,
    width=1024,
    guidance_scale=4.5,
    num_inference_steps=28,
).images[0]

image.save("flux_rs_sample.png")
```

### With SHARP

For dynamic resolution promotion and the paper-aligned inference pipeline, use the official SHARP codebase:

```bash
python run_sharp.py \
  --ckpt_path /path/to/FLUX-RS \
  --prompt "A satellite image of a rural market town with dense shop blocks, a bus station, surrounding crop fields, narrow feeder roads, and mixed residential and commercial parcels." \
  --width 1024 \
  --height 1024
```

## License

This checkpoint is derived from `black-forest-labs/FLUX.1-dev`. Please follow the license and usage terms associated with the base model when using or redistributing FLUX-RS.

## Citation

If you use FLUX-RS or SHARP in research, please cite:

```bibtex
@misc{zhao2026sharpspectrumawarehighlydynamicadaptation,
      title={SHARP: Spectrum-aware Highly-dynamic Adaptation for Resolution Promotion in Remote Sensing Synthesis},
      author={Bingxuan Zhao and Qing Zhou and Chuang Yang and Qi Wang},
      year={2026},
      eprint={2603.21783},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.21783},
}
```