How to set it up for offline use
While online, run one on-device upscale to download and cache the engine files. That first load fetches the 5 MB model and the inference runtime (roughly 15 to 25 MB on disk, depending on whether your browser uses WebGPU or WebAssembly), and a service worker stores them. From then on the page opens from cache, even in airplane mode. Before relying on offline use, reload this page online, turn Wi-Fi off, and test another image.
To keep a permanent shortcut, add the page to your home screen (phones) or create a shortcut from the browser's menu (desktop). It opens from cache like a small offline app.
What happens if the cache is cleared
Clearing site data removes the cached files; the next online on-device upscale downloads them again. Model files are versioned by filename, so a deploy never breaks a cached copy, and if a runtime file ever fails to load the engine repairs its own cache and retries.
Offline versus a desktop app
Desktop upscalers such as Upscayl (free, open source) and Topaz Gigapixel (paid) also run locally and offer larger scale factors, batch folders and, for Gigapixel, face restoration. This page is for the case where you do not want to install anything: one image, 4×, done in the tab you already have open.
Questions, answered
Does it work in airplane mode?
Yes, once an online on-device upscale has cached the 5 MB model and runtime. Load it, switch to airplane mode, drop an image.
Which browsers support offline use?
Any modern browser with service workers: Chrome, Edge, Firefox and Safari. WebGPU (recent Chrome, Edge and Safari) is about 6× faster than the WebAssembly fallback, but both work offline.
How much storage does it take?
About 5 MB for the model plus 15 to 25 MB for the inference runtime, so under 40 MB in total. Your browser may keep temporary image data for recovery. Signing in saves your results online in your library.
Curious what "AI upscaling" actually does? Read the plain-English guide, or go back to the upscaler.