To enlarge a photo, load it into the tool, choose a scale and a mode, and press Enhance. Standard mode finishes in about a second; AI super-resolution mode takes longer but brings back finer detail. Every step runs inside your browser — your image is never sent to a server.
At fit-to-screen size the difference is nearly invisible. Press 100% to view actual pixels — that is where the change in the edges becomes obvious.
| Standard mode | AI super-resolution | |
|---|---|---|
| How it works | Numerical enlargement | A learned model running inside your browser |
| Scales | 1.5 / 2 / 3 / 4x | 2 / 4x |
| Time | About a second | Seconds to tens of seconds, plus a one-time model download (7.7 MB for 2x, 50 MB for 4x) |
| Best for | Soft photos (blur recovery is available), large images, when you are in a hurry | Recovering fine detail in photos that are already in focus, especially at 4x |
| Requires | Any browser | A WebGPU-capable browser (Chrome, Edge, Safari 18+) |
The model is fetched once and served from the browser cache afterwards. Because it can render plausible detail that was not in the original, use standard mode when accuracy matters (evidence photos, measurement images and the like).
This solves the blur backwards instead of just boosting edges. It helps soft photos, but on a photo that is already sharp it makes things worse: the faint noise in any JPEG gets amplified too, and even a gentle setting lowered PSNR from 26.78 dB to 25.45 dB in our tests. The default is off. Raise it only for soft photos, while watching the comparison view.
This is how many times the blur-recovery calculation repeats, so it has no effect while blur recovery is off. Higher settings work harder and take longer, and the time estimate updates whenever you change a setting.
A last pass that tidies up the edges. Unlike blur recovery, this only emphasises detail that is already there. Turning it up makes an image look crisper while making it slightly less faithful, so the default of about 15% is a good place to stay.
We shrank a 1600px photo, saved it as a JPEG (quality 88), enlarged it back, and measured how closely the result matched the original. We used JPEG rather than lossless input because that is what real photos are. Higher numbers mean a closer match.
| Condition | Browser | Standard mode | AI super-resolution |
|---|---|---|---|
| 2x PSNR (dB) | 26.62 | 26.71 | 26.59 |
| 2x SSIM | 0.8312 | 0.8377 | 0.8413 |
| 4x PSNR (dB) | 23.24 | 23.21 | 24.02 |
| 4x SSIM | 0.6316 | 0.6381 | 0.6665 |
At 2x the two modes are roughly level; at 4x the AI pulls clearly ahead. The more that has been lost, the more its knowledge of what photos usually look like helps to fill the gap.
On a deliberately blurred JPEG enlarged 2x, blur recovery in standard mode raised PSNR from 23.22 dB to 24.21 dB. Even so, enlarging does not create information in any mode, and the original's fine detail never comes back in full. That is a limit no amount of processing removes.
No. Everything happens in your browser. Standard mode keeps working after you disconnect; only the AI model and the AVIF encoder are fetched on first use.
You get both. Standard mode finishes in about a second; AI super-resolution runs a learned model inside your browser and raised PSNR from 23.24 dB to 24.02 dB on JPEG photos enlarged 4x. The model is fetched once (7.7 MB for 2x, 50 MB for 4x) and cached. In both modes your image never leaves your device.
Mild blur improves clearly with blur recovery in standard mode (+1.0 dB in our tests). Severe shake or badly missed focus will not come back.
The default is off, and off is correct for photos that are already in focus — JPEG noise gets amplified too, and even a gentle setting lowered PSNR from 26.78 dB to 25.45 dB. Raise it only for soft photos, while watching the comparison view.
AI super-resolution needs WebGPU. Where the browser does not support it, the option is disabled automatically and standard mode remains available. Recent Chrome and Edge, and Safari 18 or later, support it.
Up to 16 megapixels of output. In standard mode the scale is reduced automatically beyond that; in AI mode the scale is fixed, so the page tells you it cannot run.
Yes — no sign-up, no watermark, no attribution required. The sample photos on the page come from Unsplash and are only there so you can test the tool.