No. 05

Image Upscaler

Increase an image's resolution 2× or 4× using an AI upscaling model that runs on your device.

Uploaded 0 bytes Loads an upscaling model (~20–60MB depending on scale) once, then works offline. (network monitor unsupported in this browser)

How your data is handled: everything above runs inside this browser tab; your file is never uploaded. Loads an upscaling model (~20–60MB depending on scale) once, then works offline.

AI upscalers are useful exactly where the source is precious — old photos, product shots, scans — which is also exactly the content you’d least want sitting on someone else’s server. This one runs an ESRGAN-class model locally via ONNX Runtime Web.

How to use it

  1. Drop an image.
  2. Choose 2× or 4×.
  3. Click Upscale — a progress readout tracks the model as it runs, GPU-accelerated where available.
  4. Compare with the before/after view, then download the result.

Tips & edge cases

  • This model is trained for photographic detail — it sharpens edges and textures well but isn't a substitute for source resolution that was never captured.
  • 4× takes roughly 3-4x longer than 2× and uses proportionally more memory, since it's effectively upscaling in two passes.
  • Heavily compressed source JPGs upscale their existing compression artifacts too — the model can't recover detail that was already lost.

FAQ

Is my image uploaded to a server to upscale it?
No — the upscaling model runs locally via onnxruntime-web. Nothing is sent over the network during processing.
What's the difference between 2× and 4×?
4× produces four times the linear resolution (16× the pixel count) of the source and takes proportionally longer to compute.
Does upscaling work on illustrations and screenshots, not just photos?
It's tuned for photographic detail; flat illustrations and UI screenshots can look over-sharpened. For those, a simple resize may look cleaner.