AI Image Upscaler

Two of the requests in the built-in instruction are not about size at all: reduce noise, and preserve natural skin tones and textures. That makes this page the enlargement pass to reach for when the file is noisy or compressed rather than merely small, since the clean-up and the enlargement are asked for in the same run.

Noise gets settled before detail is pushed

Higher resolution on a noisy file mostly means larger noise, so the instruction pairs the two: upscale, then reduce what grain and compression block are sitting on top of the picture. Where the source is a low-quality JPEG that has been re-saved a few times, look at the flat areas — a sky, a wall, a shadow — for the mottled blocks that a detail pass can turn into a smeared wash. Where the source has genuine grain, decide whether you want it removed or kept, and say so in the prompt.

Skin is the honest test for this pass

The instruction names skin tones specifically, and skin is where an image upscaler over-corrects in full view. Judge it on the areas where you know what should be there: the edge of an iris, individual hairs, a stubbled jaw, a freckle pattern. If those have gone smooth and the surrounding colour has shifted warmer or flatter, the pass has traded a person for a finish. Compare against the upload at the size you intend to use, checking edges, faces, textures, and small text there rather than in the preview.

Keeping the frame while the file grows

Nothing here repositions your subject, so the framing you upload is the framing you get back, and the work is confined to how the pixels read. Upload JPEG, JPG, PNG, or WEBP, keep the original file unaltered, and treat the enlarged output as a derivative you will want to check at full zoom before it goes anywhere.