What is inpainting?

Inpainting takes an existing image plus a mask marking a region, and regenerates only what is inside the mask. Crucially, the model sees the unmasked surroundings while it works, so it matches the lighting, perspective, colour and texture of what is already there. That context is why a well-executed inpaint is invisible, and why it is fundamentally different from pasting something in.

It is the standard fix for the failures that diffusion models reliably produce. Hands with the wrong number of fingers, a face that came out subtly wrong, a stray object in the background, a piece of garbled text: mask it, describe what should be there instead, regenerate. Rather than rerolling the whole image and losing a composition you liked, you repair the one part that failed.

Outpainting is the same mechanism pointed outward. The mask covers new canvas beyond the original edges, and the model extends the scene into it. This is how a square image becomes a wide banner, or a tight portrait gains room above the head for a title. It works best in small increments; extending far in one pass tends to drift away from the original scene.

Dreamify is text to image only and does not offer inpainting. If masked editing is a regular part of your workflow, a canvas-based tool such as Leonardo AI, Adobe's generative fill in Photoshop, or a local Stable Diffusion setup is what you want. Dreamify is the better fit when you are producing images from scratch rather than repairing them.