Research map / Image generation / Image restoration
LucidFlux: Caption-Free Universal Image Restoration via a Large-Scale Diffusion Transformer
Song Fei, Tian Ye, Lujia Wang, Lei Zhu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add caption-free SigLIP semantic alignment (ablation) | LucidFlux dual-branch conditioner · RealLQ250 | MUSIQ: 61.582 -> 62.000 | Improved |
| add timestep- and layer-adaptive modulation (TLCM) (ablation) | LucidFlux · RealLQ250 | MUSIQ: 62.000 -> 65.500 | Improved |
| scale to curated large-scale high-quality data (ablation) | LucidFlux · RealLQ250 | MUSIQ: 65.500 -> 73.0088 | Improved |
| fully fine-tune the Flux backbone instead of freezing it (ablation) | LucidFlux · RealLQ250 | MUSIQ: 73.01 -> 63.30 | Hurt |
| update only attention layers instead of freezing backbone (ablation) | LucidFlux · RealLQ250 | MUSIQ: 73.01 -> 66.27 | Hurt |
| use GT captions instead of caption-free SigLIP (ablation) | LucidFlux (Flux.1) · RealSR | MUSIQ: 70.2005 -> 70.1654 | No clear effect |
| use VLM captions instead of caption-free SigLIP (ablation) | LucidFlux (Flux.1) · RealSR | NIQE: 4.2893 -> 4.6249 | Hurt |
| use VLM captions instead of caption-free (ablation) | LucidFlux (Flux.1) · RealSR | added latency (s): +0.012 -> +10.057s | Hurt |
| keep top 10% by CLIP-IQA instead of 20% (ablation) | LucidFlux data curation · RealLQ250 | MUSIQ: 73.0088 -> 69.8807 | Hurt |
| relax flat-region filter to 600 and 60% (ablation) | LucidFlux data curation · RealLQ250 | MUSIQ: 73.0088 -> 70.2861 | Hurt |
| stricter blur range 200-6000 instead of 150-8000 (ablation) | LucidFlux data curation · RealLQ250 | MUSIQ: 73.0088 -> 71.3748 | Hurt |
| use LucidFlux instead of DiffBIR | diffusion restoration · DIV2K-Val | PSNR: 20.0389 -> 15.4393 | Hurt |
Later papers that compared with this method
1 later accepted paper reports LucidFlux as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- LucidNFT: LR-Anchored Multi-Reward Preference Optimization for Flow-Based Real-World Super-Resolution (NeurIPS 2026 · better on 6 of 6 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SeeSR — SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 2024) · better on 6 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MoE-DiffIR: Task-customized Diffusion Priors for Universal Compressed Image Restoration (ECCV 2024 · similarity 0.90)
- ReFIR: Grounding Large Restoration Models with Retrieval Augmentation (NeurIPS 2024 · similarity 0.88)
- DreamClear: High-Capacity Real-World Image Restoration with Privacy-Safe Dataset Curation (NeurIPS 2024 · similarity 0.88)
- Adapting Text-to-Image Generation with Feature Difference Instruction for Generic Image Restoration (CVPR 2025 · similarity 0.88)
- Sharing Key Semantics in Transformer Makes Efficient Image Restoration (NeurIPS 2024 · similarity 0.87)
- Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
