AI Video/Image Upscaler (free) — ×2 and ×4 upscaling right in your browser

Bottom Line Up Front (BLUF): The upscaler makes your video or image 2 or 4 times larger: a small frame turns into a big one, and the Real-ESRGAN neural network redraws the details. Everything runs on your own device — the file is never uploaded, no tokens are charged and there is no paid API behind it. The model runs right in the browser, so you see the result while it is being made, and the finished video downloads with the original frame rate and sound.

What the upscaler does

The upscaler takes your file and raises its resolution: ×2 or ×4. It is not just stretching the picture — the Real-ESRGAN neural network looks at the frame and redraws the details: edges become crisp, fine textures and hair stop turning into mush, and noise and compression artifacts are reduced.

The tool accepts both a video and a single image. For a video every frame is upscaled on its own and then everything is assembled back into a clip with the original frame rate and the original sound. For an image the result is a PNG at the higher resolution.

Everything runs on your device

The file is never sent to our server or anyone else's. The Real-ESRGAN weights (BSD-3 licence) load into the browser once and are then served from cache, and the computation runs on your graphics card through WebGPU. If WebGPU is unavailable, a CPU fallback kicks in — slower, but it works everywhere.

No tokens are charged for this model: there is no paid request at all. Video frames are decoded with WebCodecs, each one is upscaled locally, and the result is assembled into MP4 (H.264 + AAC) or WebM (VP9 + Opus), whichever your system supports.

Models and scale

There are two sets of weights. General — for photos and ordinary video: ×2 (RealESRGAN_x2plus) and ×4 (RealESRGAN_x4plus). Anime — for drawings, anime and cartoons: ×4 (realesr-animevideov3). A separate anime ×2 model does not exist, so “Anime ×2” is computed with the ×4 model and then carefully reduced to ×2 — the final resolution is honestly twice the original.

Tile size is a technical setting. The frame is cut into squares and each one is computed separately: a smaller tile uses less memory (important on a phone), a larger tile is faster on a powerful graphics card. There are no seams between tiles: the edges are taken with a margin and then cropped.

Before / After comparison and progress

For an image the result is shown with a Before / After slider: drag it to see the difference between the original and the upscaled version. For a video the preview updates while processing — you can watch the frames appear one by one, and the progress bar shows how many frames are already done.

A long run can be cancelled with a button. The finished file downloads with one click: an image as PNG, a video as MP4 or WebM.

Speed and limits

Speed depends on the file size, the chosen scale and whether you have WebGPU. On the CPU a long clip takes a while: a short clip may take a few minutes. Start with an image or a short fragment, look at the result, then take a bigger file.

Video needs a modern browser with WebCodecs — Chrome or Edge. The result is limited by the browser's technical ceiling: if the upscaled frame would exceed 8192 pixels on the long side, the tool says so honestly and suggests ×2 or a smaller file.

How to get the best result

The upscaler works best on frames with clear detail: faces, text, architecture, nature. Heavily compressed and noisy sources are cleaned up too, but no miracles are promised: if the detail is gone from the frame, the network will invent it in its own way.

For anime and drawings choose the Anime set — it does not smooth the outlines and does not “photograph” the line work.

Frequently asked questions

Is the file really never uploaded?

Yes. The model and its weights load into the browser and the computation runs on your device. Videos and images are never sent anywhere, and there is no paid API in this model.

How much does it cost?

Nothing. No tokens are charged for the upscale: the processing runs on your computer, not on our server.

Are the sound and the frame rate kept?

Yes. The video is assembled back at the original FPS and the audio track is carried over into the result (AAC for MP4, Opus for WebM). If your browser cannot encode audio, the tool says honestly that the sound could not be carried over.

What is the difference between ×2 and ×4?

×4 raises the resolution four times on each side (that is 16 times the area) and gives more detail, but takes longer and needs more memory. ×2 is faster and lighter, good for a phone.

Can I upscale an image?

Yes, the tool accepts images too: the result is a PNG at the higher resolution, with a Before / After slider.

Why is the processing slow?

Speed depends on the file size, the scale and whether WebGPU is available. On the CPU a long clip takes several times longer — try a short clip or an image.

Similar models