Spectral Forcing makes pixel-space diffusion more efficient by filtering noisy high frequencies over time, improving ImageNet (JIT) and T2I benchmark (SenseNova-U1) without adding parameters.</p>\n","updatedAt":"2026-06-17T02:49:39.407Z","author":{"_id":"6481764e8af4675862efb22e","avatarUrl":"/avatars/fc2e076bc861693f598a528a068a696e.svg","fullname":"weichenfan","name":"weepiess2383","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":7,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6841064095497131},"editors":["weepiess2383"],"editorAvatarUrls":["/avatars/fc2e076bc861693f598a528a068a696e.svg"],"reactions":[],"isReport":false}},{"id":"6a32e6b8950c6c21b4276738","author":{"_id":"6960eca92f7ad9b043b5cbe0","avatarUrl":"/avatars/e68dcc7fd04f143d849d40414866e633.svg","fullname":"Noah","name":"noahml","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":0,"isUserFollowing":false},"createdAt":"2026-06-17T18:26:00.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Cool paper. The idea of using Spectral Forcing to explicitly handle the signal-to-noise frequency split rather than letting the denoiser figure it out on its own is pretty clever. It makes a lot of sense that the model shouldn't waste capacity trying to model noise in the high-frequency bands during the early stages of diffusion.\n\nI'm curious, since the operator is parameter-free, do you think there's any risk of losing fine-grained details if the cutoff expands too aggressively?\n\nI made a podcast on it with ResearchPod, it makes it easy to get the key concepts on the go:\nhttps://researchpod.app/episode/b88b636d-e715-4fd0-bce7-608ecb9237ef","html":"<p>Cool paper. The idea of using Spectral Forcing to explicitly handle the signal-to-noise frequency split rather than letting the denoiser figure it out on its own is pretty clever. It makes a lot of sense that the model shouldn't waste capacity trying to model noise in the high-frequency bands during the early stages of diffusion.</p>\n<p>I'm curious, since the operator is parameter-free, do you think there's any risk of losing fine-grained details if the cutoff expands too aggressively?</p>\n<p>I made a podcast on it with ResearchPod, it makes it easy to get the key concepts on the go:<br><a href=\"https://researchpod.app/episode/b88b636d-e715-4fd0-bce7-608ecb9237ef\" rel=\"nofollow\">https://researchpod.app/episode/b88b636d-e715-4fd0-bce7-608ecb9237ef</a></p>\n","updatedAt":"2026-06-17T18:26:00.789Z","author":{"_id":"6960eca92f7ad9b043b5cbe0","avatarUrl":"/avatars/e68dcc7fd04f143d849d40414866e633.svg","fullname":"Noah","name":"noahml","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":0,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.944070041179657},"editors":["noahml"],"editorAvatarUrls":["/avatars/e68dcc7fd04f143d849d40414866e633.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.15236","authors":[{"_id":"6a3209a6bc818ff14e453dbc","user":{"_id":"6481764e8af4675862efb22e","avatarUrl":"/avatars/fc2e076bc861693f598a528a068a696e.svg","isPro":false,"fullname":"weichenfan","user":"weepiess2383","type":"user","name":"weepiess2383"},"name":"Weichen Fan","status":"claimed_verified","statusLastChangedAt":"2026-06-17T11:21:08.252Z","hidden":false},{"_id":"6a3209a6bc818ff14e453dbd","name":"Haiwen Diao","hidden":false},{"_id":"6a3209a6bc818ff14e453dbe","name":"Penghao Wu","hidden":false},{"_id":"6a3209a6bc818ff14e453dbf","name":"Ziwei Liu","hidden":false}],"publishedAt":"2026-06-16T00:00:00.000Z","submittedOnDailyAt":"2026-06-17T00:00:00.000Z","title":"Show the Signal, Hide the Noise: Spectral Forcing for Pixel-Space Diffusion","submittedOnDailyBy":{"_id":"6481764e8af4675862efb22e","avatarUrl":"/avatars/fc2e076bc861693f598a528a068a696e.svg","isPro":false,"fullname":"weichenfan","user":"weepiess2383","type":"user","name":"weepiess2383"},"summary":"Pixel-space diffusion models are trained on full-bandwidth noisy images, yet the useful signal available to the denoiser is strongly frequency dependent. Under rectified-flow diffusion and natural-image power-law spectra, the per-band data-to-noise contour k^{*}(t) = (1-t)^{-2/α} separates a signal-bearing low-frequency region from a noise-dominated high-frequency region at each time t. We show that this implicit coarse-to-fine structure is not merely descriptive: it induces a capacity-allocation problem. A standard pixel-space denoiser must discover the moving bandwidth boundary internally and can spend computation on frequency-time regions where the optimal prediction collapses to deterministic baselines rather than data-distribution modeling. To make this boundary explicit, we introduce Spectral Forcing, a parameter-free, time-conditional 2D-DCT low-pass operator applied to the noisy input before the patch embedder. Its cutoff expands monotonically with the diffusion time and becomes the identity at the data endpoint. Through controlled synthetic experiments, we identify the regime in which the operator is beneficial: coarse patch tokenization and data whose high-frequency content is predominantly noise rather than essential signal. On ImageNet-256 with JiT-700M/32, Spectral Forcing consistently improves both FID and Inception Score across different training epochs, demonstrating robust gains throughout training; at finer tokenization, the spectral forcing is still competitive. We further insert the unchanged operator into SenseNova-U1, a unified text-to-image model, where it improves DPG-Bench and GenEval, showing that the input-side spectral prior transfers beyond class-conditional generation. These results suggest a route to capacity-efficient pixel-space diffusion by showing the signal and hiding the noise.","upvotes":15,"discussionId":"6a3209a7bc818ff14e453dc0","githubRepo":"https://github.com/WeichenFan/Spectral_Forcing","githubRepoAddedBy":"user","ai_summary":"Spectral Forcing, a time-conditional 2D-DCT low-pass operator, improves diffusion model efficiency by explicitly separating signal from noise in pixel-space models.","ai_keywords":["diffusion models","pixel-space","denoiser","frequency-dependent","rectified-flow diffusion","power-law spectra","data-to-noise contour","capacity-allocation problem","spectral forcing","2D-DCT","patch embedder","FID","Inception Score","ImageNet-256","JiT-700M/32","DPG-Bench","GenEval","text-to-image model","senseNova-U1"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":16,"organization":{"_id":"62d55f243bf5e059f7ca25ba","name":"mmlab-ntu","fullname":"MMLab@NTU","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1658151991971-62b5777f593a2c49da69dc02.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6481764e8af4675862efb22e","avatarUrl":"/avatars/fc2e076bc861693f598a528a068a696e.svg","isPro":false,"fullname":"weichenfan","user":"weepiess2383","type":"user"},{"_id":"6565bc5ee5aac326bfc98e39","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/vIfHy9Y1yAK6A96UCHNBH.jpeg","isPro":false,"fullname":"Ting Pan","user":"PhyscalX","type":"user"},{"_id":"65af6f6b52e1b2aae437af2e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65af6f6b52e1b2aae437af2e/sFC98zLL_ZPS9fvZFi01W.jpeg","isPro":false,"fullname":"Ziang Cao","user":"Caoza","type":"user"},{"_id":"62ab1ac1d48b4d8b048a3473","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1656826685333-62ab1ac1d48b4d8b048a3473.png","isPro":false,"fullname":"Ziwei Liu","user":"liuziwei7","type":"user"},{"_id":"652d06833b5997ed71ce5c46","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/652d06833b5997ed71ce5c46/O_D6bpa5mGxLA7uCjmVCG.jpeg","isPro":false,"fullname":"Zhongang Cai","user":"caizhongang","type":"user"},{"_id":"667b8de7a68bf81afe668afe","avatarUrl":"/avatars/aeff10805ff858332e6f6a58735dbbd9.svg","isPro":false,"fullname":"leoli","user":"lifuguan","type":"user"},{"_id":"66d347eebb76fb26eedb256e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66d347eebb76fb26eedb256e/iCPF7GkmZu--XCsWzoucl.jpeg","isPro":false,"fullname":"tianqi liu","user":"tqliu","type":"user"},{"_id":"66aa94cbd59743aa4a65646f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66aa94cbd59743aa4a65646f/Z6BKgT_oYcfJgGHQNm9K3.png","isPro":false,"fullname":"Runmao Yao","user":"yaorunmao","type":"user"},{"_id":"643379416c6ecd58798421b3","avatarUrl":"/avatars/831db7eab2663abc33b176cf386b02f2.svg","isPro":false,"fullname":"Zhuoran Jin","user":"jinzhuoran","type":"user"},{"_id":"64b4a717aa03b6520839e9b8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64b4a717aa03b6520839e9b8/Rt3ERG-6BVEA4hAwOz0_I.jpeg","isPro":false,"fullname":"Haiwen Diao","user":"Paranioar","type":"user"},{"_id":"6658d01c6f1a71ba56d6c273","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/tc4nZrMuZQLfgt5aVxtH4.jpeg","isPro":false,"fullname":"Tian Shulin","user":"shulin16","type":"user"},{"_id":"63c5d43ae2804cb2407e4d43","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1673909278097-noauth.png","isPro":false,"fullname":"xziayro","user":"xziayro","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"62d55f243bf5e059f7ca25ba","name":"mmlab-ntu","fullname":"MMLab@NTU","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1658151991971-62b5777f593a2c49da69dc02.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.15236.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2606.15236 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2606.15236 in a Space README.md to link it from this page.