AI Image Tools

Image tools split cleanly by what you are trying to do, and picking the wrong type is the most common reason people conclude the whole category is disappointing. Generation makes something from nothing. Editing changes something that already exists. Upscaling and restoration recover detail. A tool that is excellent at one is frequently mediocre at the others, and the marketing rarely distinguishes.

For generation, the practical differences between tools are control and consistency rather than raw quality — most current models produce a good image from a good prompt. What separates them is whether you can get the same character, product or style twice, and whether you can steer composition without rewriting the prompt from scratch. Tools offering reference images, regional control or seed reuse are solving the problem people actually have, which is not "make one nice picture" but "make forty that belong together".

For editing, the thing to test is how the tool handles the boundary. Removing an object, extending a background or changing a colour is easy in the middle and hard at the edges, and edge quality is what determines whether the result survives being looked at closely. Try it on a photo with hair, foliage or a reflection before deciding.

Licensing deserves more attention than it usually gets. Terms vary widely on commercial use, on whether outputs can be used in a logo or a product, and on what happens to your rights if you cancel. If the images are going anywhere near a paying customer, read the terms rather than assuming the permissive default.

Cost is usually per-image or per-credit, which makes it easy to reason about and easy to underestimate, because the real number is images generated rather than images kept. A workflow that takes ten attempts to get one usable result costs ten times what the pricing page implies. Tools with fast, cheap draft modes and an expensive final render handle this well.

There is a growing set of tools that run entirely on your own machine. They are slower and need a capable GPU, and in exchange nothing leaves your computer and the marginal cost is zero — which for high-volume iteration can flip the economics completely. See [local models](/categories/local-models) for that side of the category.