Most “AI image upscaler” tools share the same catch: you upload your image to someone else’s server, wait, and download a bigger version. That’s fine for a one-off hero image. It’s useless when you’re scanning a hundred MLS thumbnails, triaging product photos, or skimming a gallery — and it means every image you touch leaves your machine.

Ultra Zoom now does the opposite. When you hover a low-resolution image, it can reconstruct detail on the fly, entirely on your device. No upload, no queue, no server that sees what you’re looking at.

What “upscaling” actually means here

Browser zoom and the hover overlay both enlarge an image — they stretch the same pixels, so a tiny thumbnail just gets blurrier the more you magnify it. Upscaling is different: it reconstructs plausible detail that wasn’t in the enlarged bitmap, using a model trained on millions of image pairs.

Ultra Zoom uses Real-ESRGAN (the general x4v3 model) to do a 4× reconstruction at hover time. When the model isn’t available for a given image, it falls back automatically to Lanczos resampling — still sharper than naive stretching, with no failure state you have to think about.

Where it earns its keep

On-device upscaling is most useful exactly where source images are small and you can’t get a bigger original:

  • MLS and listing thumbnails — real-estate portals often serve compressed previews. Upscaling makes it easier to spot a cracked soffit or a water stain before you book a showing. (More on that workflow in our real-estate listing photo guide.)
  • Compressed product photos — marketplace listings frequently ship images that pixelate the moment you zoom. Reconstructed detail helps you read a label or inspect a seam.
  • Old or archival images — scanned documents and low-resolution archive scans gain legibility.
  • Any tiny preview — avatars, gallery tiles, search-result thumbnails.

It is not magic: upscaling can’t invent text that was never captured, and a heavily artifacted JPEG has limits. But for the “I just need to see this a little more clearly” case, it’s a step change over plain magnification.

Why on-device matters

Running the model in the browser isn’t only a privacy nicety — it changes what the feature can be.

  1. Zero network calls. The model is bundled with the extension. Like everything else in Ultra Zoom, no image data is sent anywhere. This is the same zero-knowledge architecture the rest of the extension follows.
  2. No per-image latency tax. There’s no upload round-trip, so upscaling can happen at hover time instead of being a deliberate “process this file” action.
  3. Works on images you can’t re-host. Behind-login MLS photos, internal dashboards, anything you wouldn’t (or legally couldn’t) upload to a third-party upscaler.

Cloud upscalers can’t make those three promises at once. That’s the whole reason we shipped it locally. We compare the trade-offs in detail on our AI image upscalers page.

Turning it on

On-device AI upscaling is a Pro feature, with a per-site toggle so you can enable it only where you need it — the popup remembers your choice per domain. That keeps the model off the pages where plain magnification is already fine, and on for the sites where reconstruction actually helps.

It shipped as part of our recent release wave alongside listing-aware overlays and a batch of new site plugins — see the full what’s new log. If you want to try it, grab Ultra Zoom for Chrome or Firefox; Pro is a 7-day free trial when you’re ready — see plans.