Inpainting

Inpainting is what happens when image generation gets applied to editing rather than creation from scratch. Instead of generating a whole new image, you tell the model exactly which pixels to replace. A user typically draws a mask over the region (a person in the background, an unwanted object, a sky that needs to change) and provides a text prompt describing what should go in that spot. The model synthesizes new content that matches the surrounding context in lighting, perspective, and style.

Adobe popularized the consumer version of this as 'Generative Fill' in Photoshop, where you can remove objects and have the background automatically filled in, or add new elements into a scene. Google's Nano Banana models also do heavy-duty inpainting, letting you make targeted edits with natural language. Most serious image generation tools now include inpainting as a core feature.

The hard part of inpainting is seamlessness: the filled region needs to match the surrounding image in texture, lighting, and coherence. Modern models have gotten very good at this, to the point where removing a person from a photo or replacing a product in an advertising image is routine production work, not a Photoshop skill that requires expertise.