Hugging Face Daily Papers · May 18, 2026 · 4 min read

FashionChameleon: Towards Real-Time and Interactive Human-Garment Video Customization

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

Like Read original ↗

<video src=\"https://cdn-uploads.huggingface.co/production/uploads/63818941f496d57325c661c7/e5gICpyOlxpt4IEORTypk.mp4\" controls=\"\" class=\"max-w-full!\"></video></p>","updatedAt":"2026-05-18T03:07:17.905Z","author":{"_id":"63818941f496d57325c661c7","avatarUrl":"/avatars/2fd3f2846041f7273ccd18b4810f52d0.svg","fullname":"shen","name":"DukeShen","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.48262450098991394},"editors":["DukeShen"],"editorAvatarUrls":["/avatars/2fd3f2846041f7273ccd18b4810f52d0.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.15824","authors":[{"_id":"6a0a7aea75184a0d71e0263f","name":"Quanjian Song","hidden":false},{"_id":"6a0a7aea75184a0d71e02640","user":{"_id":"63818941f496d57325c661c7","avatarUrl":"/avatars/2fd3f2846041f7273ccd18b4810f52d0.svg","isPro":false,"fullname":"shen","user":"DukeShen","type":"user","name":"DukeShen"},"name":"Yefeng Shen","status":"claimed_verified","statusLastChangedAt":"2026-05-18T07:46:42.420Z","hidden":false},{"_id":"6a0a7aea75184a0d71e02641","name":"Mengting Chen","hidden":false},{"_id":"6a0a7aea75184a0d71e02642","name":"Hao Sun","hidden":false},{"_id":"6a0a7aea75184a0d71e02643","name":"Jinsong Lan","hidden":false},{"_id":"6a0a7aea75184a0d71e02644","name":"Xiaoyong Zhu","hidden":false},{"_id":"6a0a7aea75184a0d71e02645","name":"Bo Zheng","hidden":false},{"_id":"6a0a7aea75184a0d71e02646","name":"Liujuan Cao","hidden":false}],"publishedAt":"2026-05-15T00:00:00.000Z","submittedOnDailyAt":"2026-05-18T00:00:00.000Z","title":"FashionChameleon: Towards Real-Time and Interactive Human-Garment Video Customization","submittedOnDailyBy":{"_id":"63818941f496d57325c661c7","avatarUrl":"/avatars/2fd3f2846041f7273ccd18b4810f52d0.svg","isPro":false,"fullname":"shen","user":"DukeShen","type":"user","name":"DukeShen"},"summary":"Human-centric video customization, particularly at the garment level, has shown significant commercial value. However, existing approaches cannot support low-latency and interactive garment control, which is crucial for applications such as e-commerce and content creation. This paper studies how to achieve interactive multi-garment video customization while preserving motion coherence using only single-garment video data. We present FashionChameleon, a real-time and interactive framework for human-garment customization in autoregressive video generation, where users can interactively switch garment during generation. FashionChameleon consists of three key techniques: (i) Instead of training on multi-garment video data, we train a Teacher Model with In-Context Learning on a single reference-garment pair. By retaining the image-to-video training paradigm while enforcing a mismatch between the reference and garment image, the model is encouraged to implicitly preserve coherence during single-garment switching. (ii) To achieve consistency and efficiency during generation, we introduce Streaming Distillation with In-Context Learning, which fine-tunes the model with in-context teacher forcing and improves extrapolation consistency via gradient-reweighted distribution matching distillation. (iii) To extend the model for interactive multi-garment video customization, we propose Training-Free KV Cache Rescheduling, which includes garment KV refresh, historical KV withdraw, and reference KV disentangle to achieve garment switching while preserving motion coherence. Our FashionChameleon uniquely supports interactive customization and consistent long-video extrapolation, while achieving real-time generation at 23.8 FPS on a single GPU, 30-180times faster than existing baselines.","upvotes":52,"discussionId":"6a0a7aea75184a0d71e02647","projectPage":"https://quanjiansong.github.io/projects/FashionChameleon/","githubRepo":"https://github.com/quanjiansong/FashionChameleon","githubRepoAddedBy":"user","ai_summary":"FashionChameleon enables real-time interactive multi-garment video customization through teacher-student distillation and in-context learning techniques while maintaining motion coherence.","ai_keywords":["autoregressive video generation","in-context learning","streaming distillation","gradient-reweighted distribution matching","kv cache rescheduling","garment switching","motion coherence","real-time generation","single-garment video data","teacher-student distillation"],"githubStars":60,"organization":{"_id":"64488b334988ee01f2a8d856","name":"alibaba-inc","fullname":"alibaba-inc","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/61ac8f8a00d01045fca0ad2f/MX4wxQVaFm1A1wqnrL2WU.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"63818941f496d57325c661c7","avatarUrl":"/avatars/2fd3f2846041f7273ccd18b4810f52d0.svg","isPro":false,"fullname":"shen","user":"DukeShen","type":"user"},{"_id":"661f9a1b142a51d630dc165d","avatarUrl":"/avatars/b0d5325c4ce7cb632bc22766ebbecdec.svg","isPro":false,"fullname":"whalelin_Lin","user":"whalelin","type":"user"},{"_id":"68639a200b262058aa4b44cc","avatarUrl":"/avatars/b3f94124fcfd0a3d86e9db1da9a2b608.svg","isPro":false,"fullname":"Hao Sun","user":"sunhao242","type":"user"},{"_id":"633407947eb49986ce070a6c","avatarUrl":"/avatars/84245495d36f605a900950a3a76d4386.svg","isPro":false,"fullname":"song yiren","user":"songyiren","type":"user"},{"_id":"68583220f979cc06bbd13072","avatarUrl":"/avatars/f3710e9c45cceb4464eb2e54b86251a9.svg","isPro":false,"fullname":"Stefano Real","user":"stayreal1994","type":"user"},{"_id":"66fbb64725d0b1bf22bf0559","avatarUrl":"/avatars/7a0cffc7d15c755ff23a1452f32aa41d.svg","isPro":false,"fullname":"Cora Combe","user":"cocome","type":"user"},{"_id":"664b72f049350d5628fe623d","avatarUrl":"/avatars/94ad6b9895239259391199a1c610235c.svg","isPro":false,"fullname":"Absolution of Sinner","user":"abslab","type":"user"},{"_id":"66d9e56a6ef2523463e8864d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66d9e56a6ef2523463e8864d/8iEz_3dYogZxaf3M3tq1c.jpeg","isPro":false,"fullname":"AI Coder","user":"Codestral","type":"user"},{"_id":"66b4e706ff493858f554a3bc","avatarUrl":"/avatars/9140dfb293e84c9d6dd3f4025be713e7.svg","isPro":false,"fullname":"自助协力","user":"zumulun","type":"user"},{"_id":"689cc39fa5e276a382cf150e","avatarUrl":"/avatars/cfb3da86c3bea3cf627530792e7b6b65.svg","isPro":false,"fullname":"Xing Yangwei","user":"xing139565","type":"user"},{"_id":"68ccc02e22940ad759036eb1","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/68ccc02e22940ad759036eb1/YA_DNs9dfUgCLoNJWJZ6Q.png","isPro":false,"fullname":"Xqy","user":"Kytolly","type":"user"},{"_id":"664b82a78e6575f45ed9623e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/664b82a78e6575f45ed9623e/4sB0F1rObsAHHtDlLqpZx.jpeg","isPro":false,"fullname":"Prefecture","user":"DoubleTV","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"64488b334988ee01f2a8d856","name":"alibaba-inc","fullname":"alibaba-inc","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/61ac8f8a00d01045fca0ad2f/MX4wxQVaFm1A1wqnrL2WU.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.15824.md"}">

Papers

arxiv:2605.15824

FashionChameleon: Towards Real-Time and Interactive Human-Garment Video Customization

Published on May 15

· Submitted by

shen on May 18

alibaba-inc

Upvote

Authors:

Yefeng Shen ,

Abstract

FashionChameleon enables real-time interactive multi-garment video customization through teacher-student distillation and in-context learning techniques while maintaining motion coherence.

AI-generated summary

Human-centric video customization, particularly at the garment level, has shown significant commercial value. However, existing approaches cannot support low-latency and interactive garment control, which is crucial for applications such as e-commerce and content creation. This paper studies how to achieve interactive multi-garment video customization while preserving motion coherence using only single-garment video data. We present FashionChameleon, a real-time and interactive framework for human-garment customization in autoregressive video generation, where users can interactively switch garment during generation. FashionChameleon consists of three key techniques: (i) Instead of training on multi-garment video data, we train a Teacher Model with In-Context Learning on a single reference-garment pair. By retaining the image-to-video training paradigm while enforcing a mismatch between the reference and garment image, the model is encouraged to implicitly preserve coherence during single-garment switching. (ii) To achieve consistency and efficiency during generation, we introduce Streaming Distillation with In-Context Learning, which fine-tunes the model with in-context teacher forcing and improves extrapolation consistency via gradient-reweighted distribution matching distillation. (iii) To extend the model for interactive multi-garment video customization, we propose Training-Free KV Cache Rescheduling, which includes garment KV refresh, historical KV withdraw, and reference KV disentangle to achieve garment switching while preserving motion coherence. Our FashionChameleon uniquely supports interactive customization and consistent long-video extrapolation, while achieving real-time generation at 23.8 FPS on a single GPU, 30-180times faster than existing baselines.