Drop an image
or click to choose · paste works too Ctrl V
How it works
- 1
Add an image
Drop, paste or pick a JPG, PNG or WebP.
- 2
Choose 2x or 4x
The panel shows the size you will get before you spend anything.
- 3
Compare and download
Drag the divider, hold still for the zoom lens, then save the PNG.
Detail, not blur
Why SwinIR looks different from a resize.
Skin, fabric and foliage are reconstructed from what the model learned, not smeared.
JPEG blocks and grain are cleaned in the same pass.
A before/after divider and a 3x zoom lens show exactly what changed.
PNG download at the full size, no watermark.
What model does the upscaler use?
SwinIR, a Swin Transformer trained for real-world photos (the BSRGAN-degraded ‘real SR’ model). It rebuilds detail and removes compression noise instead of only stretching pixels.
How big can the result be?
Choose 2× or 4×. Images up to 1600 pixels on the long side go in, so a 4× result can reach 6400 pixels across.
Will it work on AI images, screenshots and old photos?
Yes. It is strongest on photos and AI pictures. Text in screenshots stays readable but may look slightly painted; for faces in old photos, try Restore old photos, which rebuilds faces too.
Is there a watermark?
No. You download the full-resolution PNG with nothing added.
How much does it cost?
Three free runs without an account, then a fixed number of credits per image whatever its size. No subscription.
- 01
Passport photo maker
Any selfie, sized to your country's rules, with a print sheet.
30+ countries→ - 02
Restore old photos
Faces rebuilt, scratches softened, black and white turned to colour.
GFPGAN + DDColor→ - 03
Object remover
Paint over what should go. The gap is filled in.
LaMa→ - 04
Vocal remover
Split a song into vocals and instrumental, or clean up a voice.
Demucs→