AI upscaler

Resizing stretches the pixels you already have, so a small picture becomes a large blurry one. Upscaling reconstructs what should be there: the model has seen enough hair, brick, fabric, and foliage to infer the detail that the original file never recorded, and it draws that back in as it enlarges. The result is a bigger image with more information, not the same information spread thinner.

Photographs, artwork, screenshots, scans, and video. It is most useful on the files you cannot reshoot: an old family photo, a logo that only survives as a small PNG, a clip that was compressed to nothing by a messaging app, or footage shot on an older camera that now has to sit in a modern timeline.

Comfortably several times the original dimensions, which takes most web-sized images up to print or full-resolution use. What matters more than the multiplier is what the source contains. A clean small image upscales beautifully. A heavily compressed one has artefacts baked in, and enlarging it enlarges those too, so start from the best copy you can find.

No. There are no sliders to balance and no denoise-versus-sharpen tradeoff to guess at. Upload the file and the model handles reconstruction. If you want a different character in the result, switch models rather than hunting for the right combination of settings.

Generally yes, and this is where model choice earns its keep. Faces and small type are the two things that expose a weak upscaler, because both are places where a viewer knows immediately when the detail is wrong. Run an important file through two models and keep whichever holds the eyes and the lettering better.

Yes. Video upscaling rebuilds every frame and has to keep the result stable across them, so detail does not shimmer or crawl from one frame to the next. That temporal consistency is the hard part, and it is why video uses models built for it rather than an image upscaler run frame by frame.

Yes. Save it as a Workflow and run each file through the same treatment. A product catalogue or an archive of scans comes out uniform, which matters more than any single file when the set will be seen together.