Upscales an image 4× with an anime-tuned AI model (Real-ESRGAN) on your device — nothing is uploaded. Best for small illustrations / anime art. Large inputs are scaled down first; the first run loads the model.

Drop an image here

or click to choose · or paste (Ctrl/Cmd + V)

processed in your browser · never uploaded

Drop or click to replace

When you want a small illustration, anime artwork, or icon bigger and crisper, this AI image upscaler runs entirely on your device. Drop an image in (or click to choose, or paste) and press Upscale 4× — an in-browser AI model (the anime-tuned Real-ESRGAN) raises the resolution 4×, and you can download the result. Unlike plain enlargement (nearest / bilinear), it keeps lines and flat areas cleaner as it scales up. Importantly, your image is never uploaded, stored, or sent anywhere: the AI inference runs entirely inside your browser, so unreleased art or personal images stay on your device. The model and runtime load only on first use (about 10 MB in total) and run from cache afterwards. It uses the full anime-tuned Real-ESRGAN model (anime_6B), so edges stay sharp. On supported browsers it uses the GPU (WebGPU) for speed, otherwise WASM. To protect memory and speed, inputs are scaled to about 512px on the long side before upscaling (larger images are scaled down first). Because the model is tuned for line art and anime-style images, photographs and complex textures may not upscale as well.

How to use

Drop an image in, click to choose, or paste (Ctrl/Cmd + V).
Press Upscale 4× (the first run loads the model, so it takes a moment).
When the result appears, Download PNG. Your image is never sent anywhere.

FAQ

Is my image uploaded to a server?

No. All processing, including the AI inference, runs in your browser (onnxruntime-web / wasm / WebGPU). Your image is never uploaded, stored, or sent — it stays on your device.

What images work best?

It's tuned for anime-style and line-art illustrations and icons, especially smaller images. Photographs and complex textures may not upscale as well.

It's slow or heavy.

Upscaling is compute-heavy AI inference. On WebGPU-capable browsers it's much faster; otherwise it runs on WASM and takes longer. Inputs are limited to about 512px on the long side before the 4× (up to 2048px).

Are large images upscaled as-is?

To protect memory and speed, large inputs are scaled down to about 512px on the long side first, then upscaled 4×. It suits enlarging small illustrations more than re-rendering large photos.

Why is the first run slow?

It loads the AI model and runtime once (about 10 MB in total). After that they're cached in your browser and skip re-downloading.