# Stream-DiffVSR: Low-Latency Streamable Video Super-Resolution via Auto-Regressive Diffusion

Source: [Stream-DiffVSR: Low-Latency Streamable Video Super-Resolution via Auto-Regressive Diffusion](<https://www.sota2.com/research/paper/69535c8c53fdd5c36173d9f6>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-08T05:00:04.658Z

Authors: Hau-Shiang Shiu, Chin-Yang Lin, Zhixiang Wang, Chi-Wei Hsiao, Po-Fan Yu, Yu-Chih Chen, Yu-Lun Liu

Published: 2025-12-29T18:59:57.000Z

## Abstract

Diffusion-based video super-resolution (VSR) methods deliver strong perceptual quality but are often unsuitable for latency-sensitive scenarios due to reliance on future frames and expensive multi-step denoising. We propose Stream-DiffVSR, a causally conditioned diffusion framework for efficient online VSR. Operating strictly on past frames, Stream-DiffVSR integrates a four-step distilled denoiser for fast inference, an Auto-regressive Temporal Guidance (ARTG) module that injects motion-aligned cues during latent denoising, and a lightweight temporal-aware decoder with a Temporal Processor Module (TPM) to enhance detail and temporal coherence. Unlike chunk-wise streaming inference, our strictly frame-by-frame causal design avoids sequence-level waiting, substantially reducing time-to-first-frame and end-to-end latency. Stream-DiffVSR processes 720p frames in 0.328 seconds on an RTX 4090 and consistently outperforms prior diffusion-based baselines. Compared with the online state-of-the-art TMP, it improves perceptual quality (LPIPS +0.095) while reducing latency by over 130x. Moreover, Stream-DiffVSR substantially lowers time-to-first-frame for diffusion-based VSR, reducing initial delay from over 4600 seconds to 0.328 seconds, making diffusion-based VSR markedly more practical for low-latency online and streaming deployment. Project page: https://jamichss.github.io/stream-diffvsr-project-page/

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Video Super-Resolution on VideoLQ (NIQE, NRQM, BRISQUE)](<https://www.sota2.com/research/sota/video-super-resolution-on-videolq-niqe-nrqm-brisque.md>): Video Super-Resolution; VideoLQ; NIQE: 3.929.

- [Video Super-Resolution Performance (Runtime/Memory) on Vimeo-90K-T](<https://www.sota2.com/research/sota/video-super-resolution-performance-runtime-memory-on-vimeo-90k-t.md>): Video Super-Resolution; Vimeo-90K-T; Runtime Per Frame (s): 0.67.

- [Image-to-Video Generation on VBench-I2V (2K resolution)](<https://www.sota2.com/research/sota/image-to-video-generation-on-vbench-i2v-2k-resolution.md>): Image-to-Video Generation; VBench-I2V 2K resolution; Total Score: 6.424.

- [Image-to-Video Generation on 81-frame 2K video 2560 × 1408 (Efficiency)](<https://www.sota2.com/research/sota/image-to-video-generation-on-81-frame-2k-video-2560-1408-efficiency.md>): Image-to-Video Generation; 81-frame 2K video 2560 × 1408; Peak Memory Usage (GB): 49.2.

- [Video Super-Resolution on VideoLQ (test) (NRQM, NIQE, PI, BRISQUE)](<https://www.sota2.com/research/sota/video-super-resolution-on-videolq-test-nrqm-niqe-pi-brisque.md>): Video Super-Resolution; VideoLQ (test); NRQM: 6.14.

- [Video Super-Resolution on REDS4 (test)](<https://www.sota2.com/research/sota/video-super-resolution-on-reds4-test.md>): Video Super-Resolution; REDS4 (test); PSNR (Avg): 27.256.

- [Video Super-Resolution on Vid4 (test)](<https://www.sota2.com/research/sota/video-super-resolution-on-vid4-test.md>): Video Super-Resolution; Vid4 (test); PSNR: 22.725.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2512.23709>)

[PDF](<https://arxiv.org/pdf/2512.23709.pdf>)