AI Image Upscaler
Enlarge & sharpen images 4× with on-device AI — no upload.
Runs entirely in your browser — nothing you enter is uploaded, logged, or tracked.
A real AI image upscaler that runs 100% on your device — drop in a small, soft or low-resolution image and an on-device super-resolution neural network (no servers, no upload) reconstructs it at up to 4× the size with sharper edges and recovered detail, not a blurry stretch (very large images are first scaled to 1024 px on the long edge, for an output up to ~4096 px). Choose a Photo model for photographs or an Illustration model for drawings, anime and artwork, pick 2× or 4×, then drag the divider to compare the AI result against a plain enlargement. Export a PNG, JPEG or WebP. The models download once (a few MB) and then work offline forever; your images never leave your device, and the export carries no GPS or camera metadata.
Export
Lossless still image.
Re-encoded on your device — no GPS or camera metadata, no watermark.
Frequently asked questions
Is my image uploaded anywhere?
No. The AI super-resolution model and all processing run inside your browser on your own device — there are no servers, no upload, no analytics. After the one-time model download you can disconnect from the internet and it still works. Cloud upscalers send your images to their servers; this one never does.
How does an AI upscaler run in my browser with nothing uploaded?
It uses a compact super-resolution neural network (a SPAN / Real-ESRGAN-family model) compiled to run on your device via WebAssembly. The first time you use it the model downloads once (just a few MB), caches, and from then on every upscale is computed locally and offline. Large images are processed in overlapping tiles so it works without running out of memory.
Is this real detail or just a bigger blurry image?
It’s real reconstruction — the model was trained to predict plausible high-resolution detail (sharper edges, cleaner textures) rather than just stretching pixels, so the result looks far crisper than a normal resize. It can’t invent information that was never there, so extremely tiny or heavily compressed images have limits, but for most small or soft photos the improvement is dramatic. Drag the divider to see the AI result vs a plain enlargement.
What’s the difference between the Photo and Illustration models?
The Photo model is tuned for photographs (faces, scenes, textures). The Illustration model is tuned for drawings, anime, line art and flat-colour artwork, where it keeps edges clean and avoids photo-style noise. Try both and keep the one that looks best.
How big can I upscale, and how long does it take?
It upscales up to 4×, with the input capped at 1024 px on the long edge (so up to ~4096 px out). Time depends on the image size and your device — small images take a few seconds, larger ones tens of seconds, with a progress bar per tile. It all runs on your CPU in the browser, so it’s slower than a cloud GPU but completely private.
Does it remove EXIF / GPS metadata?
Always. The export is re-encoded from raw pixels, so the original’s GPS location, camera model and EXIF tags can’t carry over — there’s nothing to toggle.