FlashVSR 1.1 image model

FlashVSR upscales low-res or blurry videos fast. You send the clip, it bumps up the resolution and sharpens the video.

FlashVSR v1.1 has been released in November 2025. OpenImagingLab, part of Shanghai AI Lab, teamed up with researchers from Tsinghua, CUHK, and Shanghai Jiao Tong. They're working on solving the ongoing problem of using diffusion models for video upscaling that actually works in real-world cases.

You can scale your clips to 720p, 1080p, 2K or 4K. Just upload and wait. It's said to beat most tools on output quality and keeps each frame more stable. 

It improves video quality because it smooths out flicker and ghosting so the motion stays steady. Brings back fine textures like hair or fabric without going too sharp. Cleans up blocky compression junk and ringing. Handles quick movement without messing up pans or fast scenes. Keeps things crisp but not fake-looking.

They ran tests on a single A100 GPU, probably with 40 to 80 GB of VRAM, and hit around 17 fps at 768×1408 resolution. One example showed it used about 11.1 GB of memory, while a similar setup needed over 52 GB. So if you're running this locally, plan for a GPU with at least 11 GB for medium resolutions. For higher ones, you'll need a lot more.