4K upscaler – Upscale images to 4K

Up to 4× its size. A photo of at least 960 px on the long side reaches 4K (3840 px); a smaller one is enlarged 4× and the page shows the size you get, because stretching it further would only add blur.

An AI model trained on millions of photo pairs (small and large) looks at every part of your picture and draws it again with 2 or 4 times the pixels. Unlike simple enlarging, which only blurs between the existing pixels, it rebuilds edges, lines and fine texture, and removes JPG blocks and noise on the way.

No upscaler can recover detail that the photo never captured. Natural mode stays close to what is there. Enhance details and Max add realistic texture such as pores, leaves or fabric weave that the AI invents: it looks convincing, but it is not what the camera saw. Don't use those modes for documents, evidence or identifying people.

Tiny faces (about 20 px), text smaller than about 6 px, and photos that are strongly blurred or very heavily compressed. No model in this class recovers those; they stay soft or come out invented. The best results come from small but clean photos, 300 to 1500 px.

With a graphics chip (WebGPU in Chrome, Edge or Safari) a typical photo takes seconds. Without one the AI runs on the processor: a 1-megapixel photo takes about a minute in Natural mode and several minutes with Enhance details. The page says when a mode will be slow on your device.

Natural: waifu2x swin_unet photo models by nagadomi (MIT licence), 10 MB. Enhance details: 4xBHI_realplksr_dysample_real and Max: 4xBHI_dat2_real, both by Philip Hofmann (CC BY 4.0), 15 and 29 MB. They are downloaded once, kept on your device and run in your browser.

No. The AI model is downloaded to your browser and runs there; your photos never leave your device. That is also why the first use takes a moment longer: the model is fetched once and then kept for later visits.

Yes, with no account, no watermark and no limit on the number of files. FotoJock is made by Foto Jocković, a photo shop in Belgrade.