AI Image Upscaler

2x is the default and covers most jobs. Doubling twice is usually worse than doubling once — each pass works from the previous guess rather than from the original.

It should not. Crop, colour and composition stay put; only resolution moves. If you want the picture itself changed, that is the editor.

Soft, yes. Motion-blurred, no — camera shake is missing information rather than soft information, and enlarging it produces a larger blur.

Yes, and often better than on photographs: line art and brush texture have cleaner edges to work from. The illustrated pair above is the example.

Shown before you run it, drawn from the same balance as everything else, and refunded if the run fails.

After. Edit at draft size, settle the direction, then enlarge the keeper once.

A plain resize interpolates: it averages neighbouring pixels to invent the ones in between, which is why enlarged pictures go soft. An upscaler was trained on what real detail looks like at both sizes, so it reconstructs edges rather than averaging across them. The difference is most visible on hair, fabric weave and small text — exactly the places averaging fails first.

It should not, and that is the property worth checking on any upscaler you use. Some models reconstruct so aggressively that a face comes back subtly different — a stranger with your subject's haircut. Compare the pairs above at full size before trusting one with portrait work.

Yes — four times the pixels at 2x, and roughly proportionate file size. Worth knowing before you upscale a whole catalogue. Enlarge the pictures that will actually be seen large, and leave the thumbnails alone.

You can, but the second pass works from the first pass's guesses rather than from real detail, so the returns fall off sharply. One pass at the size you need beats two passes stacked.

Ones that are sharp but small. A well-focused photograph at 800 pixels has real detail in every edge, and enlarging gives that detail room. A soft picture at 3000 pixels has less to work with despite being larger — sharpness matters more than starting size.

On text rendered large enough to be legible, yes — letterforms have precise edges and reconstruct cleanly. Text that is already too small to read comes back larger and still unreadable, because the shapes were never captured in the first place.