Advanced Image Techniques: Inpainting, Outpainting, and ControlNet

Part of the Prompt Engineering course

In the last two lessons you generated whole images from a text prompt and learned to steer them with structure, style, and negative prompts. That approach gives the model wide creative latitude, which is a strength when you are exploring and a problem when you need a specific result.

In lesson eight a generation began from random noise and denoised toward your prompt. Image-to-image (often written img2img) changes the starting point: instead of noise, the model begins from an existing image and denoises from there. The prompt still guides the result, but the source picture pulls it toward a familiar shape. ...