The slider is a choice, not a quality setting
Push it towards repair and you get a clean, sharp, confident face that is less and less the face in the photograph. Push it the other way and you keep the person, along with more of the damage.
There is no correct position. For a stranger in a crowd, repair freely. For somebody you know, stay close — a relative’s face rebuilt into a generically attractive one is worse than the blurry original, and everyone who knew them can tell.
When to use the other one instead
This tool only touches faces and leaves the picture the same size. If you also want the whole photograph enlarged and cleaned up, the photo restore tool does both in one pass.
For a landscape, a product or anything without a face, neither is right — the upscaler is.
A photo of a face, sent to a server
A picture of someone’s face deserves a plain answer about where it goes, so: it is uploaded. It is one of the AI tools here that does; the rest of the site works inside your browser.
It travels from your browser straight to storage on a signed address that expires in five minutes, is read once, and is deleted with the result after about three hours. It is not kept, not looked at, and not attached to any account — there are no accounts.
Frequently asked questions
Will it look like the same person?
At a low repair setting, largely yes. At a high one, no — it produces a plausible face rather than recovering the real one, because the information simply is not in the pixels. This is the single most important thing to understand about the tool.
Does it work on several faces at once?
Yes, every face it finds is repaired. A group photo where one face is sharp and another is blurred will have both processed, so check the sharp one still looks right afterwards.
Is it free?
Yes, with no account and no watermark. The only cap is on how often one network can use it in an hour, so that one script cannot take the machine away from everyone else.
What size photo can I send?
Up to five megabytes, in JPG, PNG, WebP, GIF or BMP. A face occupying a small part of a large photo is worth cropping first — the model works on what it is given, and giving it mostly background wastes the detail budget on scenery.
Something wrong with this tool, or an idea for it? Tell us