Is this AI upscaler really free?

Yes, without limits. Online upscalers charge credits because the work runs on their servers; here it runs on your own computer, so it costs nothing to offer and there is nothing to meter.

How long does upscaling take?

Seconds for a small image on a computer with a modern graphics chip, and a minute or two for a larger one on the processor alone - 4× takes several times longer than 2×. The progress bar counts the tiles and estimates the time left.

Can it restore a blurry old photo?

It sharpens edges and adds plausible fine detail, which helps a small or softly focused photo a great deal, and the 4× photo model is trained for exactly that. It cannot recover what was never captured - a face that is only a few pixels wide stays a guess - and the added detail is invented, so it is not suitable where exact fidelity matters.

Why is there a size limit?

Every input pixel becomes four or sixteen output pixels, and each one is worked out by the model, so time and memory grow quickly. Images over about 4 megapixels are scaled down first; cropping to the part you need gives the model more to work with.

Stays on your computer Downloads an AI model from huggingface.co the first time: about 7 MB for 2x and 21-22 MB for the 4x modes, or, where WebGPU is used, about 8 MB and 53-55 MBHow this works

Free AI image upscaler

Enlarge a small or blurry picture 2× or 4× with an AI model that adds real detail instead of smudging it — free, unlimited, and on your own computer

Drag & drop an image here, or browse

JPG, PNG, WebP, GIF, AVIF or HEIC · works best up to about 1000 px · nothing is uploaded

About this AI image upscaler

Enlarging a picture the ordinary way only spreads the same pixels further apart, so it goes soft and blocky. A super-resolution model has learned from thousands of high-resolution photos, shrunk and compared with the originals, what fine detail usually looks like - the texture of skin, the edge of a letter, the weave of fabric - and fills it in. This uses Swin2SR, a well-regarded open model, in three versions: a fast 2× one, a 4× one trained on real camera photos with their noise and blur, and a 4× one that also repairs the blocky squares of a heavily compressed JPEG.

The model runs on your computer - on the graphics chip where the browser supports WebGPU, otherwise on the processor - so there are no credits to buy, no daily limit and no watermark. The image is processed in small overlapping tiles, which keeps memory in check and leaves no seams, and a transparent PNG keeps its transparency. It is free to use, with no account and no upload limit, and it runs entirely in your browser: nothing is uploaded, and while the page stays open, a feature you have already used keeps working offline.

What you can do

Upscale an image 2× or 4× with an AI model that adds detail instead of blurring it
A 4× mode trained on real camera photos, for noisy or soft pictures
A 4× mode that also cleans up the blocky squares of a heavily compressed JPEG
Opens JPG, PNG, WebP, GIF, AVIF and iPhone HEIC images
Best on pictures up to about 1000 pixels across; anything over 4 megapixels is scaled down first
Drag a slider across the result to compare it with the original
Save the result as PNG or JPG
Transparent PNGs keep their transparency
Runs on the graphics chip where the browser supports WebGPU, otherwise on the processor
A tile-by-tile progress bar with an estimate of the time left
No watermark and no daily limit

How to upscale an image

Drop the picture onto the page, or click to choose it.
Pick 2× for speed, 4× for photos, or 4× repair for a blocky JPEG. The model downloads the first time.
Press Upscale, then drag across the result to compare it with the original, and save it.

Questions

Is this AI upscaler really free?: Yes, without limits. Online upscalers charge credits because the work runs on their servers; here it runs on your own computer, so it costs nothing to offer and there is nothing to meter.
How long does upscaling take?: Seconds for a small image on a computer with a modern graphics chip, and a minute or two for a larger one on the processor alone - 4× takes several times longer than 2×. The progress bar counts the tiles and estimates the time left.
Can it restore a blurry old photo?: It sharpens edges and adds plausible fine detail, which helps a small or softly focused photo a great deal, and the 4× photo model is trained for exactly that. It cannot recover what was never captured - a face that is only a few pixels wide stays a guess - and the added detail is invented, so it is not suitable where exact fidelity matters.
Why is there a size limit?: Every input pixel becomes four or sixteen output pixels, and each one is worked out by the model, so time and memory grow quickly. Images over about 4 megapixels are scaled down first; cropping to the part you need gives the model more to work with.