Private by construction
There is no server to upload to, so the file never leaves your machine.
Nothing uploaded →
Runs entirely on your GPU
AI Upscale video and images for free
A super-resolution network runs locally through WebGPU. Your file never leaves the browser — no upload, no account, no watermark and no limit.
- Your Media
- Your GPU
- Your Privacy
Move your cursor to scrub
Rendering preview…
Zoom in to judge the detail — scroll to zoom, drag to pan, drag the divider to compare.
Preparing…
Keep this tab in the foreground — background tabs are throttled.
Everything runs on your machine
No server, no queue, no upload. The whole pipeline — decode, inference, encode — happens inside this tab.
Real GPU inference
WGSL compute shaders through WebGPU — not a slow CPU fallback.
Hardware accelerated →Audio kept intact
Audio is copied through untouched, so sync and quality both survive.
No re-encoding →Preview before you commit
Upscale a single frame and zoom to 8×. It is the real output.
Zoom and compare →Four steps, start to finish
-
01
Choose a file
Drop in a video or image. It is read locally — no upload happens.
-
02
Compare and tune
Pick a resolution and model, then zoom in to judge the detail.
-
03
Upscale
The network runs over every frame on your GPU and writes a new file.
-
04
Download
Save the result. No watermark, no limit, nothing kept.
Pick your trade-off
Read the FAQFast
8,612
Eight channels. Best for long clips and modest GPUs.
Balanced
31,036
Sixteen channels. The sensible default for most files.
Quality
90,592
Twenty-eight channels. For detail-critical work.
Good to know
Is my file uploaded anywhere?
No. There is no backend. The page is a static bundle and everything runs inside the tab — you can disconnect from the network after it loads and it still works.
What do I need to run it?
A browser with WebGPU and WebCodecs — Chrome, Edge or Opera 113 and newer — on a machine with a GPU. The badge in the header tells you whether you are good to go.
How much can it enlarge?
The networks output exactly 2×. Lower rungs such as 1080p come from resampling that result. Anything beyond 2× would be interpolation rather than reconstruction, so it is not offered.
Which formats work?
Any video your browser can decode — typically H.264 MP4, WebM and MOV — plus PNG, JPEG and WebP images. Output is MP4 for video and PNG for stills. HDR and 10-bit video are not supported.
Why is it slower on some files?
Every frame runs through the network on your GPU, so time scales with resolution, length and the model you pick. Large images are processed in tiles to bound memory use. Keep the tab in the foreground — browsers throttle background tabs.
Is the upscaler really free?
Yes. No sign-up, no credits, no paid tier and no cap on how many files you run. The work happens on your own GPU, so there is no server bill to pass on to you, and the whole thing is open source on GitHub.
Does the output have a watermark?
No. You get a plain MP4 or PNG with nothing overlaid, at the resolution you picked, and you are free to use it anywhere.
Does it work offline?
Yes, once the page and the model have loaded. There is no backend to call, so you can disconnect from the network and keep upscaling. The browser caches the bundle, so a return visit starts faster.
Which model does it use?
Anime4K-style super-resolution networks compiled to WGSL and run through WebGPU, in three sizes — fast at eight channels, balanced at sixteen and quality at twenty-eight. A wider network reconstructs more detail per frame and costs more time per frame.
Can I use it on a phone or tablet?
Only where the browser exposes both WebGPU and WebCodecs, which on mobile is still patchy. Desktop Chrome, Edge or Opera on a machine with a real GPU is the reliable path today.
Is there a file size or length limit?
None is imposed by the tool. In practice your GPU memory and your patience are the limits — large images are processed in tiles so memory stays bounded, and long videos simply take longer.