Motion blur

For handheld camera shake, the traditional starting point is a shutter speed at least as fast as one over the focal length in 35 mm equivalent terms, so roughly 1/50 of a second at a 50 mm equivalent view. Treat that as a rough guideline rather than a rule: it assumes a steady hand, and higher-resolution sensors show small wobbles that older cameras hid.

Subject movement needs its own budget, independent of the lens. A hand adjusting a garment, a pet, or a hanging item still swinging after you positioned it will all need a faster shutter than the guideline suggests. Image stabilisation, whether in the lens or the body, counters camera movement only; it does nothing for a subject that is moving on its own.

For still products the cheapest fix is not a faster shutter at all. Put the camera on a tripod, use the self-timer or a remote so pressing the button does not shake it, and then a slow shutter costs you nothing.

Not properly. Sharpening raises contrast along whatever edges survived, so applied to a streaked photo it gives you a bolder streak. Deconvolution, which tries to reverse the smear mathematically, can help when the blur is short, uniform and in a single known direction, but real handheld shots usually blur along a small curve rather than a straight line, and any noise in the file gets amplified as the algorithm works backwards.

Machine learning deblurring produces the most convincing results and is the least honest about what it is doing. It reconstructs plausible detail from what it has seen before, which means the model number on a smeared label may come back crisply legible and wrong. For a listing photo where a buyer relies on what is written on the product, reshooting is both faster and safer than arguing later about a returned item.