Enlarge a picture

Increase image quality

Opens on the quality recipe: light clean-up first, detail 40 inside the halo guard, 2× — or the same size, cleaner.

Upscaled inside this tab. The original never travels.
Drop something small. We'll show you how far it can go.Paste works too, and so does a file you drag straight off the desktop — or several at once, for a batch. JPG, PNG, WebP, BMP and the HEIC an iPhone shoots open here; AVIF and TIFF open where this browser can read them (TIFF in Safari). Every one of them stays on this machine.

Nothing small to hand? There is a drawn 240 × 160 one.

Size 2× · 3× · 4× · 8×, any factor or exact pixelsPrint sizes at 150, 240 and 300 PPIEnhance on-device models, off until you ask

What this opens on

The Quality panel, with the recipe in force: light clean-up first (it softens JPEG block steps before they are enlarged), then detail at 40 with a noise floor of 3 levels so grain is left alone, and the halo guard on so no edge grows a rim. Output is 2× by default; choose Same size, cleaner to keep your pixel count.

A worked case: a 640 × 480 JPEG saved at quality 40 shows its 8-pixel block grid in flat areas. Light clean-up first removes most of that grid, so the 2× result (1,280 × 960) enlarges the picture rather than the grid.

Every pass runs inside this tab; the model, if you add one, comes to the picture — the picture never goes to it.

Where this falls down

  • Focus and motion blur stay.No label here says unblur, restore or recover, because none of these passes does that.
  • Clean-up removes texture too.Every step up softens fine detail along with the damage. Judge it at 1:1 and use the lowest level that clears the grid.
  • A model can invent the wrong thing.Faces and lettering are where trained models produce confident mistakes. Check names and numbers at 1:1 before relying on them.

Questions people ask

Can this fix a blurry photo?
No. Blur from focus or movement is information that was never recorded, and no clean-up or sharpening recovers it. What this page does is remove compression damage, steepen edges inside a halo guard, and — if you choose — blend in a trained model's proposal. The guide on still-soft pictures explains how to tell a blur from a size problem.
What does 'same size, cleaner' do?
It keeps your picture's pixel size. The clean-up and detail passes run at 1×; if a model has run, its result (made at its own 3× or 4×) is brought back down to your size with the Lanczos kernel and blended in at the strength you choose. The record says so: output factor 1, model run at 4×, brought back down.
Is the model's detail real?
No — it is invented. A trained model proposes texture that looks plausible; it has never seen your scene. That is why the Smooth result is always beside it, and why the record that travels with the file says which pixels were proposed.
Is anything uploaded?
No. The clean-up, the detail pass and the model all run in this tab. A model file is downloaded to your device only after you press a button that states its size.

Better fit?

Increase Image Quality
Photographic visual for Increase Image Quality