Batch Image Upscale

Upscaling increases the pixel dimensions of an image, and the important thing to understand first is that no upscaler recovers information that was never captured. Detail absent from the source cannot be restored, only plausibly invented or smoothly interpolated. Every honest description of the technique starts there.

Simple interpolation averages between existing pixels. The result is larger and softer, with no new detail and no invented artefacts either. It is predictable and safe, and it is what you want when the goal is simply to meet a minimum dimension requirement rather than to look better.

Model-based upscaling is different in kind. It has learned what edges, textures and faces usually look like, and it reconstructs plausible detail consistent with that training. On photographs of ordinary subjects the result can be genuinely impressive. It is also, strictly speaking, inventing: the detail is a statistically likely guess rather than a record of what was in front of the camera.

That distinction matters in specific situations and not at all in others. For a product image on a web page, invented texture is fine. For anything evidential, a document scan, a medical image, a photograph used to establish a fact, invented detail is a serious problem precisely because it looks convincing. An upscaled licence plate or a sharpened face is not evidence of anything, and treating it as such is the failure mode worth naming.

Practical limits: the further you push the factor, the more of the output is invention rather than reconstruction, and the more visible the seams become. Text and faces are where errors are most noticeable, because a reader knows exactly what those should look like. Compression artefacts in the source are amplified along with everything else, so upscaling a heavily compressed JPEG enlarges its flaws.

As with every tool here, the work happens in your browser and the images are never uploaded, which is what makes it usable for material you are not free to send elsewhere.

A common follow-up: which mode should you pick? 2x roughly quadruples the pixel count compared to the original, and 4x quadruples it again from there, so processing time and output file size grow the same way. Start at 2x for web pages, social posts and thumbnails, where the difference from 4x is rarely visible on screen. Reach for 4x when the destination is print or a large display that specifically needs the extra pixels.