Hugging Face Daily Papers · May 26, 2026 · 3 min read

ControlLight: Towards Controllable, Consistent, and Generalizable Low-Light Enhancement

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

ControlLight resources:\n<ul>\n<li>Paper: <a href=\"https://arxiv.org/abs/2605.25569\" rel=\"nofollow\">https://arxiv.org/abs/2605.25569</a></li>\n<li>Project page: <a href=\"https://yfyang007.github.io/ControlLight/\" rel=\"nofollow\">https://yfyang007.github.io/ControlLight/</a></li>\n<li>Model: <a href=\"https://huggingface.co/ControlLight/ControlLight\">https://huggingface.co/ControlLight/ControlLight</a></li>\n<li>Dataset: <a href=\"https://huggingface.co/datasets/ControlLight/Light100K\">https://huggingface.co/datasets/ControlLight/Light100K</a></li>\n<li>Code: <a href=\"https://github.com/yfyang007/ControlLight\" rel=\"nofollow\">https://github.com/yfyang007/ControlLight</a></li>\n</ul>\nControlLight supports continuous enhancement-strength control with consistent outputs while preserving image structure and natural visual details.\n","updatedAt":"2026-05-26T04:24:54.444Z","author":{"_id":"6731a7f033691aafb3dafcfc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6731a7f033691aafb3dafcfc/Ql6viYWw-T8Qg2LWmMu6a.jpeg","fullname":"Yufeng Yang","name":"dericky286","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6799843311309814},"editors":["dericky286"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/6731a7f033691aafb3dafcfc/Ql6viYWw-T8Qg2LWmMu6a.jpeg"],"reactions":[],"isReport":false}},{"id":"6a153483ab2b9b6831d2440a","author":{"_id":"6731a7f033691aafb3dafcfc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6731a7f033691aafb3dafcfc/Ql6viYWw-T8Qg2LWmMu6a.jpeg","fullname":"Yufeng Yang","name":"dericky286","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false},"createdAt":"2026-05-26T05:49:55.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"\nhttps://cdn-uploads.huggingface.co/production/uploads/6731a7f033691aafb3dafcfc/Q00rD0fhNh5smTq7B5kZV.mp4\n","html":"<video src=\"https://cdn-uploads.huggingface.co/production/uploads/6731a7f033691aafb3dafcfc/Q00rD0fhNh5smTq7B5kZV.mp4\" controls=\"\" class=\"max-w-full!\"></video>\n","updatedAt":"2026-05-26T05:49:55.776Z","author":{"_id":"6731a7f033691aafb3dafcfc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6731a7f033691aafb3dafcfc/Ql6viYWw-T8Qg2LWmMu6a.jpeg","fullname":"Yufeng Yang","name":"dericky286","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6662734746932983},"editors":["dericky286"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/6731a7f033691aafb3dafcfc/Ql6viYWw-T8Qg2LWmMu6a.jpeg"],"reactions":[{"reaction":"🔥","users":["dericky286"],"count":1},{"reaction":"🚀","users":["dericky286"],"count":1}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.25569","authors":[{"_id":"6a151deab57a1823d5708b4b","name":"Yufeng Yang","hidden":false},{"_id":"6a151deab57a1823d5708b4c","name":"Jianzhuang Liu","hidden":false},{"_id":"6a151deab57a1823d5708b4d","name":"Jisheng Chu","hidden":false},{"_id":"6a151deab57a1823d5708b4e","name":"Yuqi Peng","hidden":false},{"_id":"6a151deab57a1823d5708b4f","name":"Xianfang Zeng","hidden":false},{"_id":"6a151deab57a1823d5708b50","name":"Jiancheng Huang","hidden":false},{"_id":"6a151deab57a1823d5708b51","name":"Shifeng Chen","hidden":false}],"publishedAt":"2026-05-25T00:00:00.000Z","submittedOnDailyAt":"2026-05-26T00:00:00.000Z","title":"ControlLight: Towards Controllable, Consistent, and Generalizable Low-Light Enhancement","submittedOnDailyBy":{"_id":"6731a7f033691aafb3dafcfc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6731a7f033691aafb3dafcfc/Ql6viYWw-T8Qg2LWmMu6a.jpeg","isPro":false,"fullname":"Yufeng Yang","user":"dericky286","type":"user","name":"dericky286"},"summary":"Existing deep learning-based low-light enhancement methods are typically trained on limited datasets with single enhancement targets, which restricts their generalization ability and controllability in real-world applications. To overcome these limitations, we propose ControlLight, a controllable, consistent, and generalizable framework for low-light enhancement. We first construct a large-scale dataset of real-world degraded images with continuous illumination-strength supervision. To further ensure consistent outputs under different control strengths, we introduce a misalignment-aware weighted flow matching loss that preserves image structure across continuous enhancement strengths. ControlLight allows users to edit real-world degraded low-light images toward satisfactory enhancement results by flexibly controlling the strength while preserving visual consistency and realism. Extensive experiments show that ControlLight achieves state-of-the-art performance against existing low-light enhancement approaches while demonstrating strong continuous controllability and generalization to real-world scenarios.","upvotes":10,"discussionId":"6a151debb57a1823d5708b52","projectPage":"https://yfyang007.github.io/ControlLight/","githubRepo":"https://github.com/yfyang007/ControlLight","githubRepoAddedBy":"user","ai_summary":"ControlLight is a controllable low-light enhancement framework that uses a large-scale real-world dataset and weighted flow matching loss to ensure consistent image quality across varying enhancement strengths.","ai_keywords":["low-light enhancement","continuous illumination-strength supervision","weighted flow matching loss","image structure preservation","controllability","generalization","real-world degraded images"],"githubStars":3},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64c9bac33cfe45b07179568d","avatarUrl":"/avatars/4a8206cdb1770a8cdaae0d0a2b7b59f2.svg","isPro":false,"fullname":"Pengtao Chen","user":"PengtaoChen","type":"user"},{"_id":"64b914c8ace99c0723ad83a9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64b914c8ace99c0723ad83a9/B4gxNByeVY_xaOcjwiN1j.jpeg","isPro":false,"fullname":"Wei Cheng","user":"wchengad","type":"user"},{"_id":"64ca05b4f7f4ccb5ea6e43aa","avatarUrl":"/avatars/c909613715eaf5fd43ae6cd95ae2b9a4.svg","isPro":false,"fullname":"Charles","user":"SCFW","type":"user"},{"_id":"68b04979f64bd1f33194cbcb","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/dtsZQPHpBFnOCmgHfSeNh.png","isPro":false,"fullname":"Chan","user":"yuUnuo","type":"user"},{"_id":"6742f612924e80c3c81352d9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/Wmp_PY2t-CuJEoeCyR3NM.png","isPro":false,"fullname":"Haoling Xie","user":"HAOlingX","type":"user"},{"_id":"6731a7f033691aafb3dafcfc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6731a7f033691aafb3dafcfc/Ql6viYWw-T8Qg2LWmMu6a.jpeg","isPro":false,"fullname":"Yufeng Yang","user":"dericky286","type":"user"},{"_id":"64ad04093c80c402fbc42860","avatarUrl":"/avatars/31c97293c06da17d60b4359cabc66da2.svg","isPro":false,"fullname":"FU ZHOUJIE","user":"Kr1sJ","type":"user"},{"_id":"634bde123d11eaedd889e277","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1665916392312-noauth.png","isPro":false,"fullname":"Hengyuan Xu","user":"DobyXu","type":"user"},{"_id":"67bc119fdcc189331317d987","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/M85l-BtSyZI_ZaPBFDr--.png","isPro":false,"fullname":"JsChu","user":"JsChu","type":"user"},{"_id":"67390fd1d566e6d2a8fb61df","avatarUrl":"/avatars/7373a1e81bd10de24fa77cff04fe589a.svg","isPro":false,"fullname":"Yuqi Peng","user":"yuqii77","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.25569.md"}">

Papers

arxiv:2605.25569

ControlLight: Towards Controllable, Consistent, and Generalizable Low-Light Enhancement

Published on May 25

· Submitted by

Yufeng Yang on May 26

Upvote

Authors:

Abstract

ControlLight is a controllable low-light enhancement framework that uses a large-scale real-world dataset and weighted flow matching loss to ensure consistent image quality across varying enhancement strengths.

AI-generated summary

Existing deep learning-based low-light enhancement methods are typically trained on limited datasets with single enhancement targets, which restricts their generalization ability and controllability in real-world applications. To overcome these limitations, we propose ControlLight, a controllable, consistent, and generalizable framework for low-light enhancement. We first construct a large-scale dataset of real-world degraded images with continuous illumination-strength supervision. To further ensure consistent outputs under different control strengths, we introduce a misalignment-aware weighted flow matching loss that preserves image structure across continuous enhancement strengths. ControlLight allows users to edit real-world degraded low-light images toward satisfactory enhancement results by flexibly controlling the strength while preserving visual consistency and realism. Extensive experiments show that ControlLight achieves state-of-the-art performance against existing low-light enhancement approaches while demonstrating strong continuous controllability and generalization to real-world scenarios.