license: other
pipeline_tag: image-to-image
tags:
- image-restoration
- low-light-enhancement
- deflare
- diffusion
- nighttime-photography
- siggraph-2026
LUCID
LUCID: Learning Unified Control for Image Deflaring and Exposure Mastery in Nighttime Photography
LUCID is a unified framework for nighttime image restoration. It targets the coupled degradations that often appear together in night photography: underexposure, intense flare, ghosting artifacts, and visible light sources.
Photography is the art of painting with light, yet nighttime scenes are shaped by competing degradations: intense flares obscure scene structure, while photon-limited regions collapse into noise. Conventional approaches address these factors in isolation, overlooking the fact that these degradations are fundamentally entangled. LUCID reframes nighttime restoration as a continuous and controllable process rather than a fixed correction, restoring challenging nighttime images with flexible control over exposure, light sources, flare, and ghosting artifacts.
Checkpoints
This repository contains the pretrained checkpoints used by the LUCID codebase:
| File | Description |
|---|---|
LUCID_main/model_40000.pkl |
Main LUCID restoration checkpoint |
Flare_Disentangle/latest.pth |
Flare disentanglement network checkpoint |
The base diffusion model is SD-Turbo and should be downloaded separately:
https://huggingface.co/stabilityai/sd-turbo
Usage
Clone the code repository:
git clone https://github.com/frakenation/LUCID.git
cd LUCID
Download these checkpoints:
git lfs install
git clone https://huggingface.co/Unswear/LUCID lucid_weights
Example inference:
python -m src.inference \
--input_dir ./data/test/input \
--pretrained_model_name_or_path stabilityai/sd-turbo \
--model_path ./lucid_weights/LUCID_main/model_40000.pkl \
--flare_disentanglement_path ./lucid_weights/Flare_Disentangle/latest.pth \
--output_dir ./results/lucid \
--resolution 512 \
--timestep 199 \
--ms_unet \
--inference_mode cfg_guidance \
--cfg_scale 1.05 \
--device cuda
For multi-scale exposure control, use src.inference_cfg_index or the provided shell scripts in the GitHub repository.
Links
- Project page: https://xiaoyunyuan.net/index.html?project=lucid
- Paper: https://arxiv.org/abs/2606.06901
- Code: https://github.com/frakenation/LUCID
- Video: https://www.youtube.com/watch?v=AGPLSiZcK_I
Citation
If you find this project useful, please cite the paper from the project page or arXiv.