Scenario
← All Models
Video

Flash VSR Upscale Video

Use it ↗

Flash VSR by Tsinghua/Shanghai AI Lab upscales video 2x to 4x using one-step diffusion, delivering sharp detail restoration with real-time streaming speed.

Flash VSR is an ultra-fast one-step diffusion model designed for real-time, high-fidelity video super-resolution. By utilizing locality-constrained sparse attention and a lightweight decoder, it delivers a 12x speed boost over prior models, upscaling footage to 4K at 17 FPS while ensuring sharp textures and flawless temporal stability

More models from Academia