LiveVVT: High-Fidelity Video Virtual Try-On in Real Time

Yushe Cao; Shikun Feng; Ruxiang Duan; Liyong Wang; Dianxi Shi; Chun Yu; Junliang Xing

cs.CVAug 27, 2026

LiveVVT: High-Fidelity Video Virtual Try-On in Real Time

Authors: Yushe Cao, Shikun Feng, Ruxiang Duan, Liyong Wang, Dianxi Shi, Chun Yu, Junliang Xing

Organizations: Tsinghua University · Zhongguancun Academy · South China University of Technology · Beijing Jiaotong University

Abstract

Diffusion-based Video Virtual Try-On (VVT) achieves high visual fidelity through bidirectional spatio-temporal modeling, but complete-clip dependence incurs prohibitive latency and computational overhead in practical continuous deployment. Naively enforcing causality disrupts pretrained bidirectional priors and substantially degrades synthesis quality. We introduce LiveVVT, a rolling streaming diffusion framework that preserves bounded bidirectional modeling within causal recurrent generation. Within a fixed-size window, LiveVVT jointly denoises multiple video chunks under bounded look-ahead, preserving local bidirectional interactions while emitting one clean chunk per iteration. Beyond the window, two complementary memories sustain long-term consistency: a bounded temporal memory propagates recent dynamics and occlusion context, whereas a persistent global appearance memory, constructed once from the target garment and a frontal try-on keyframe, anchors garment details and dressed appearance throughout the stream. We further introduce a progressive distillation framework integrating bidirectional VVT learning, teacher-trajectory regression for causal few-step adaptation, and Collaborative Matching Distillation, which couples teacher-distribution matching with rolling flow matching on real videos to align optimization with recurrent inference. Experiments on paired and unpaired long-sequence benchmarks demonstrate superior generation quality over similarly sized models, with $26\times$ lower latency and $11\times$ higher throughput, enabling high-fidelity real-time streaming VVT.

Figures & tables

Figure 1: Comparison of latency and throughput between LiveVVT and state-of-the-art methods.

Figure 2: Overview of LiveVVT. (a) Rolling streaming try-on emits one clean chunk per update from a staggered-noise window with persistent global appearance memory $\mathcal{A}=(\mathcal{A}_{g},\mathcal{A}_{f})$ and temporal memory $\mathcal{H}_{k}$ . (b) Bidirectional VVT learning, teacher-trajectory regression, and CoMD progressively transfer offline bidirectional priors to few-step streaming generation.

	Paired				Unpaired
Method	VFID I $\downarrow$	VFID R $\downarrow$	SSIM $\uparrow$	LPIPS $\downarrow$	VFID I $\downarrow$	VFID R $\downarrow$
OOTDiff+AM	46.666	24.450	0.720	0.238	47.802	25.857
CatVTON+AM	45.754	20.707	0.739	0.231	46.125	20.728
VACE	18.086	0.418	0.762	0.149	27.519	0.871
ViViD	20.887	0.337	0.792	0.130	27.963	0.462
CatV2TON	20.694	0.232	0.815	0.132	28.343	0.356

Table 1: Quantitative comparison of LiveVVT with competing methods on the long-sequence ViViD-SL benchmark.

	Paired				Unpaired
Method	VFID I $\downarrow$	VFID R $\downarrow$	SSIM $\uparrow$	LPIPS $\downarrow$	VFID I $\downarrow$	VFID R $\downarrow$
OOTDiff+AM	33.225	2.311	0.756	0.246	39.139	2.172
CatVTON+AM	33.334	2.555	0.769	0.246	36.863	1.547
VACE	20.373	0.405	0.787	0.173	31.152	1.648
ViViD	24.244	0.611	0.785	0.145	28.356	1.037
CatV2TON	22.018	0.376	0.811	0.156	30.653	1.133

Table 2: Quantitative comparison of LiveVVT with competing methods on the long-sequence ViT-HDL benchmark.

Figure 3: Long-Term and Multi-View Try-On Comparison.

Panel of Figure 4 — Figure 4: Long-stream scalability: (a) per-update latency, (b) total inference time, and (c) peak GPU memory usage.

Figure 5: Qualitative comparison across progressive distillation stages under rolling inference.

Configuration	VFID I $\downarrow$	VFID R $\downarrow$	SSIM $\uparrow$	LPIPS $\downarrow$	FPS $\uparrow$
$\mathcal{A}_{g}$ only	14.415	0.131	0.824	0.094	26.74
$+\,\mathcal{A}_{f}$	13.960	0.115	0.826	0.093	25.85
$+\,\mathcal{H}$	13.871	0.112	0.832	0.088	22.60
$+\,\mathcal{H},\mathcal{A}_{f}$ (Ours)	13.194	0.090	0.838	0.099	22.39

Table 3: Ablation of the frontal try-on and temporal memory components on ViViD-SL under paired setting.

Training stage	VFID I $\downarrow$	VFID R $\downarrow$	SSIM $\uparrow$	LPIPS $\downarrow$
Stage I	17.202	0.281	0.786	0.144
Stage II	14.371	0.156	0.803	0.116
Stage III	13.194	0.090	0.838	0.099

Table 4: Ablation of the three progressive distillation stages under rolling inference on ViViD-SL under paired setting.

Appendix figures & tables12 assets

Supplementary material from the paper’s appendix.

Appendix

	Paired				Unpaired
Method	VFID I $\downarrow$	VFID R $\downarrow$	SSIM $\uparrow$	LPIPS $\downarrow$	VFID I $\downarrow$	VFID R $\downarrow$
OOTDiff+AM	42.955	1.816	0.656	0.259	47.001	2.622
CatVTON+AM	33.043	1.422	0.714	0.213	39.726	1.192
VACE	23.589	0.875	0.835	0.128	32.292	2.378
ViViD	28.712	0.812	0.791	0.149	35.954	1.316
CatV2TON	29.888	1.633	0.796	0.163	37.385	1.420

Appendix

Table 5: Full-sequence quantitative comparison on the long-sequence TikTokDress-L benchmark under paired and unpaired settings.

Figure 6: Average single-image try-on inference time of LiveVVT and state-of-the-art image try-on methods, evaluated at a resolution of

512\times 384

	Paired				Unpaired
Method	FID $\downarrow$	KID $\downarrow$	SSIM $\uparrow$	LPIPS $\downarrow$	FID $\downarrow$	KID $\downarrow$
IDM-VTON	6.338	1.322	0.881	0.079	9.611	1.639
OOTDiff	9.305	4.086	0.819	0.088	12.408	4.689
CatVTON	6.139	0.964	0.869	0.097	9.143	1.267
UniFit	8.799	0.702	0.883	0.065	–	–
Ours	5.910	0.419	0.873	0.054	9.212	0.889

Appendix

Table 6: Quantitative comparison of LiveVVT with publicly available image try-on methods on the VITON-HD test set.

$\lambda$	VFID I $\downarrow$	VFID R $\downarrow$	SSIM $\uparrow$	LPIPS $\downarrow$
0	14.515	0.205	0.826	0.094
0.1	12.807	0.123	0.834	0.096
0.2	13.194	0.090	0.838	0.099
0.4	13.435	0.146	0.838	0.104
0.8	14.446	0.146	0.843	0.104

Appendix

Table 7: Ablation of the RFM weight

\lambda

in CoMD on ViViD-SL under the paired setting. The highlighted row denotes the default configuration used in the main experiments.

Conditioning	VFID I $\downarrow$	VFID R $\downarrow$	SSIM $\uparrow$	LPIPS $\downarrow$	FPS $\uparrow$
Cached	13.194	0.090	0.838	0.099	22.39
Repeated Encoding	13.280	0.103	0.840	0.089	19.73

Appendix

Table 8: Quality and throughput comparison of garment-conditioning strategies under the paired setting. Cached stores garment KV features in the global appearance memory, whereas Repeated Encoding recomputes garment conditioning at every rolling update.

Figure 7: Long-sequence comparison for a white blouse under substantial pose and scale changes.

Figure 8: Long-sequence comparison for denim shorts across large viewpoint changes.

Figure 9: Long-sequence comparison for a dark knit top across front and rear views.

Figure 10: Long-sequence comparison for black shorts. MagicTryOn exhibits an appearance shift in later frames, whereas LiveVVT maintains consistent garment color and texture.

Figure 11: Unpaired long-sequence try-on examples across diverse garment categories and viewpoint changes. LiveVVT preserves garment appearance through full-body rotations and extended rolling generation.

Figure 12: Additional unpaired long-sequence examples with substantial viewpoint and camera-scale changes. LiveVVT maintains a coherent dressed appearance after rear views, self-occlusion, and abrupt zooming.

Figure 13: Challenging in-the-wild try-on examples with complex backgrounds, rapid motion, and severe self-occlusion. LiveVVT retains robust garment appearance and temporal continuity across these unconstrained sequences.

Virtual Try-On

Virtual try-on generates images or videos showing people wearing selected garments while preserving appearance, fit, pose, and identity.

See all 55 papers

UniVVT: A Unified End-to-End Framework for High-Fidelity Video Virtual Try-on
Aug 6, 2026Yushe Cao, Shikun Feng, Fei Shen +5Virtual Try-On Identity-Preserving Video Generation
Tsinghua University · Zhongguancun Academy · National University of Singapore +2
InstructVVT: Instruction-Driven Video Virtual Try-On without Auxiliary Spatial Priors
Aug 14, 2026Dingbao Shao, Song Wu, Xinyu Chen +18Virtual Try-On Instruction-Based Video Editing
FlowVVTON: Flow-Guided Mask-Free Video Virtual Try-On
Aug 31, 2026Shengyao Chen, Xianbing Sun, Liqing Zhang +1Virtual Try-On Temporal Consistency in Video Generation
Shanghai Jiao Tong University, Shanghai, China

Aug 6, 2026J/K move · Enter open · S save

UniVVT: A Unified End-to-End Framework for High-Fidelity Video Virtual Try-on

Yushe Cao, Shikun Feng, Fei Shen, Haikuo Peng+4

Tsinghua University · Zhongguancun Academy · National University of Singapore+2

Virtual Try-On Identity-Preserving Video Generation Controllable Video Generation

Open paper AskPDF arXiv

Video Virtual Try-On (VVT) synthesizes a video of a person wearing a target garment while preserving identity, motion, and scene dynamics. Dominant approaches cast VVT as mask-conditioned video inpainting and rely on separate modules for human parsing, pose estimation, and garment warping. This multi-stage design complicates deployment and, more critically, allows errors in explicit geometric priors to propagate irreversibly into the generated video. We present UniVVT, a unified end-to-end framework that reframes VVT as semantically conditioned video generation, eliminating mask, pose, and warping modules at inference. At its core, a scene-task perceiver built on a Multimodal Large Language Model jointly encodes the source video, target garment, and task instruction into compact, task-aware latent tokens, implicitly capturing what to transfer and where and how to transfer it. A lightweight semantic bridge then aligns these tokens with the conditioning space of a diffusion-based video generator, enabling coherent garment transfer. To robustly couple the heterogeneous components, we devise a three-stage progressive training strategy comprising semantic alignment, joint task adaptation, and flexible-resolution refinement. Extensive experiments demonstrate that UniVVT achieves state-of-the-art performance across multiple benchmarks, validating implicit semantic guidance as a simple and effective alternative to fragile geometric preprocessing for end-to-end virtual try-on.

Abstract

Virtual Try-On

Explore similar work

UniVVT: A Unified End-to-End Framework for High-Fidelity Video Virtual Try-on

InstructVVT: Instruction-Driven Video Virtual Try-On without Auxiliary Spatial Priors

FlowVVTON: Flow-Guided Mask-Free Video Virtual Try-On