LucidFlux: Caption-Free Universal Image Restoration with a Large-Scale Diffusion Transformer
† Equal contribution ‡ Project Leader * Corresponding author
Framework

Gallery
















































































































































































































Open-Source Comparison
Quantitative Results
Benchmark | Metric | Methods | |||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
ResShift | StableSR | SinSR | SeeSR | DreamClear | SUPIR | LucidFlux (Ours) |
|||||||
Caption-Free | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ | ||||||
Real-world | DRealSR | CLIP-IQA+ ↑ | 0.4655 | 0.3732 | 0.5402 | 0.6257 | 0.4461 | 0.5494 | 0.6748 | ||||
Q-Align ↑ | 2.6311 | 2.1245 | 3.1334 | 3.2745 | 2.4213 | 3.4720 | 3.6919 | ||||||
MUSIQ ↑ | 40.9795 | 29.6691 | 53.9138 | 61.3222 | 35.1911 | 54.9279 | 66.6833 | ||||||
MANIQA ↑ | 0.2687 | 0.2402 | 0.3455 | 0.4505 | 0.2675 | 0.3482 | 0.4985 | ||||||
NIMA ↑ | 4.3178 | 3.9048 | 4.6226 | 4.6401 | 3.9368 | 4.5063 | 4.9625 | ||||||
CLIP-IQA ↑ | 0.4964 | 0.3383 | 0.6631 | 0.6760 | 0.4360 | 0.5309 | 0.6879 | ||||||
NIQE ↓ | 10.3005 | 8.6022 | 6.9800 | 6.4502 | 7.0163 | 5.9091 | 4.7034 | ||||||
RealSR | CLIP-IQA+ ↑ | 0.5005 | 0.4408 | 0.5416 | 0.6731 | 0.5331 | 0.5640 | 0.7074 | |||||
Q-Align ↑ | 3.1045 | 2.5087 | 3.3615 | 3.6073 | 3.0044 | 3.4682 | 3.7555 | ||||||
MUSIQ ↑ | 49.50 | 39.98 | 57.95 | 67.57 | 49.48 | 55.68 | 70.20 | ||||||
MANIQA ↑ | 0.2976 | 0.2356 | 0.3753 | 0.5087 | 0.3092 | 0.3426 | 0.5437 | ||||||
NIMA ↑ | 4.7026 | 4.3639 | 4.8282 | 4.8957 | 4.4948 | 4.6401 | 5.1072 | ||||||
CLIP-IQA ↑ | 0.5283 | 0.3521 | 0.6601 | 0.6993 | 0.5390 | 0.4857 | 0.6783 | ||||||
NIQE ↓ | 9.0674 | 6.8733 | 6.4682 | 5.4594 | 5.2873 | 5.2819 | 4.2893 | ||||||
RealLQ250 | CLIP-IQA+ ↑ | 0.5529 | 0.5804 | 0.6054 | 0.7034 | 0.6810 | 0.6532 | 0.7406 | |||||
Q-Align ↑ | 3.6318 | 3.5586 | 3.7451 | 4.1423 | 4.0640 | 4.1347 | 4.3935 | ||||||
MUSIQ ↑ | 59.50 | 57.25 | 65.45 | 70.38 | 67.08 | 65.81 | 73.01 | ||||||
MANIQA ↑ | 0.3397 | 0.2937 | 0.4230 | 0.4895 | 0.4400 | 0.3826 | 0.5589 | ||||||
NIMA ↑ | 5.0624 | 5.0538 | 5.2397 | 5.3146 | 5.2200 | 5.0806 | 5.4836 | ||||||
CLIP-IQA ↑ | 0.6129 | 0.5160 | 0.7166 | 0.7063 | 0.6950 | 0.5767 | 0.7122 | ||||||
NIQE ↓ | 6.6326 | 4.6236 | 5.4425 | 4.4383 | 3.8700 | 3.6591 | 3.6742 | ||||||
Synthetic | DIV2K-Val | CLIP-IQA+ ↑ | 0.5583 | 0.5760 | 0.6128 | 0.7116 | 0.6585 | 0.6719 | 0.7492 | ||||
Q-Align ↑ | 3.5761 | 3.4226 | 3.7336 | 4.1167 | 3.9323 | 4.1659 | 4.5311 | ||||||
MUSIQ ↑ | 60.5932 | 57.4246 | 66.0906 | 71.4947 | 65.8187 | 67.9074 | 73.9045 | ||||||
MANIQA ↑ | 0.3421 | 0.2902 | 0.4341 | 0.5104 | 0.4369 | 0.4148 | 0.5819 | ||||||
NIMA ↑ | 5.0430 | 5.0341 | 5.1810 | 5.2709 | 5.1663 | 5.1516 | 5.4884 | ||||||
CLIP-IQA ↑ | 0.6017 | 0.5002 | 0.7166 | 0.7149 | 0.6663 | 0.5848 | 0.7034 | ||||||
NIQE ↓ | 6.1976 | 4.9810 | 5.3679 | 4.2823 | 4.1634 | 3.7701 | 3.7283 | ||||||
PSNR ↑ | 18.3802 | 18.3269 | 18.0956 | 18.2529 | 17.5701 | 17.7567 | 15.4393 | ||||||
SSIM ↑ | 0.4394 | 0.4819 | 0.4259 | 0.4684 | 0.4291 | 0.4482 | 0.3837 | ||||||
LPIPS ↓ | 0.3738 | 0.3933 | 0.3919 | 0.3497 | 0.3621 | 0.3785 | 0.4312 | ||||||
LSDIR-Val | CLIP-IQA+ ↑ | 0.5248 | 0.5576 | 0.5582 | 0.7258 | 0.6995 | 0.7126 | 0.7440 | |||||
Q-Align ↑ | 3.5317 | 3.4878 | 3.7095 | 4.2997 | 4.2391 | 4.3468 | 4.5959 | ||||||
MUSIQ ↑ | 57.6691 | 57.0838 | 63.9586 | 72.0142 | 70.7186 | 70.3340 | 74.1923 | ||||||
MANIQA ↑ | 0.3408 | 0.2990 | 0.4131 | 0.5529 | 0.5059 | 0.4482 | 0.5979 | ||||||
NIMA ↑ | 5.0916 | 5.0628 | 5.3353 | 5.4245 | 5.3773 | 5.3692 | 5.6221 | ||||||
CLIP-IQA ↑ | 0.5691 | 0.4991 | 0.6766 | 0.7314 | 0.6941 | 0.6105 | 0.6836 | ||||||
NIQE ↓ | 6.4447 | 4.2104 | 5.1771 | 3.9402 | 3.3318 | 2.9610 | 3.5571 | ||||||
PSNR ↑ | 17.3040 | 17.1480 | 16.8241 | 17.0782 | 16.2114 | 16.1598 | 14.8688 | ||||||
SSIM ↑ | 0.3935 | 0.4026 | 0.3710 | 0.4113 | 0.3823 | 0.3636 | 0.3697 | ||||||
LPIPS ↓ | 0.4824 | 0.4655 | 0.4637 | 0.3969 | 0.3720 | 0.4408 | 0.4148 |
Table 1: Quantitative comparison across different IQA metrics on RealSR, RealLQ250, DIV2K-Val, LSDIR-Val, and DRealSR datasets. Best results are highlighted in bold.
Qualitative Results
Input | SinSR | SeeSR | SUPIR | DreamClear | LucidFlux (Ours)









Commercial Comparison
Quantitative Results
Method | CLIP-IQA+ ↑ | Q-Align ↑ | MUSIQ ↑ | MANIQA ↑ | NIMA ↑ | CLIP-IQA ↑ | NIQE ↓ |
---|---|---|---|---|---|---|---|
LQ Input | 0.6218 | 2.1693 | 44.1541 | 0.3718 | 3.8664 | 0.6079 | 6.0790 |
Seedream 4.0 | 0.5002 | 3.6931 | 52.3771 | 0.2794 | 4.7024 | 0.4124 | 4.9393 |
Gemini-NanoBanana | 0.3780 | 3.3114 | 44.6310 | 0.2548 | 4.6571 | 0.4434 | 6.0865 |
MeiTu SR | 0.6653 | 4.1464 | 66.5936 | 0.4498 | 5.2103 | 0.6663 | 5.4125 |
LucidFlux (Ours) | 0.7406 | 4.3935 | 73.01 | 0.5589 | 5.4836 | 0.7122 | 3.6742 |
Table 2: Quantitative comparison across different IQA metrics with commercial models on RealLQ250. Best results are highlighted in bold.
Qualitative Results
Input | HYPIR-FLUX | Topaz | Seedream 4.0 | MeiTu SR | Gemini-NanoBanana | LucidFlux (Ours)








Contact Us
For professional collaboration and inquiries, please contact:
sfei285@connect.hkust-gz.edu.cn
tye610@connect.hkust-gz.edu.cn