AI image tools are genuinely useful and routinely oversold. The distinction that matters throughout is whether a tool is *recovering* information that survived in the file, or *inventing* information that was never there — because the second kind produces convincing results that are nonetheless fabrications.

These articles apply that lens to each capability: upscaling, denoising, enhancement, background removal, generative fill and outpainting. They also cover the ecosystem questions that follow — how to tell whether an image was generated, what Content Credentials can and cannot prove, where AI-generated images stand on copyright, and why AI-drafted alt text needs human editing before it helps anyone.