RelayVSR
by USTC (Xijun Wang et al.)
Real-world video super-resolution: upload a low-resolution clip and get it back 4× sharper, with the original audio.
Live
- Cost
- ~123 credits / run
- Time
- ~40 s per run
- License
- Apache-2.0 ✓ commercial
- Runs on
- 80-96 GB GPU shared with the other video models (Wan reference model + small Flash model)
Runs the official RelayVSR inference code and checkpoints (Wan references every 15 frames, the small Flash model in between). Up to 10 seconds per video; inputs larger than 540 px on the long side are scaled down first, so the result is at most 2160 px. The original audio is copied back. A 4-second 320×176 clip takes about 10 seconds on our 96 GB GPU. It shares the GPU with LTX-2.5 and Hy-Nomas.
Official pageCodeWeights
Questions
What is RelayVSR?
RelayVSR is an AI video model by USTC (Xijun Wang et al.). Real-world video super-resolution: upload a low-resolution clip and get it back 4× sharper, with the original audio.
How much does RelayVSR cost?
You buy a credit pack once, no subscription. A run usually costs about 123 credits (1,000 credits = US$1), charged for the real GPU time. Failed runs are refunded.
How do I use RelayVSR online?
Sign in, open this page, fill in the form and press the run button. It runs on our cloud GPUs: nothing to install, and it works on phones too.
Can I use RelayVSR results commercially?
License: Apache-2.0. Commercial use of the outputs is allowed by the license.
Where was RelayVSR presented?
In the AI Search video at 2:32.
All models · Not affiliated with USTC (Xijun Wang et al.) or with AI Search.