Image Generation

Creating new images with AI, often from text prompts or other visual input.

Does the generated still life match the brief?

The illustration on this page was generated for a specific brief: two pitchers, a tall teal one on the left and a shorter ivory one on the right. Appearance alone can distract from errors in count or position.

Count
Locate two separate pitchers. Count the vessels themselves, rather than handles, nearby decoration or shadows.
Relationship
Check left and right from the viewer’s position. Then compare the height and color of each pitcher against the brief.
Use
The image can illustrate an imagined arrangement. It cannot serve as a photograph documenting a museum’s holdings or the history of an actual vessel.

How does generation differ from image search?

A generator synthesizes an image conditioned on a prompt or other input. In a diffusion system, a learned denoising process can progressively form an image from noise. A retrieval system instead returns material already indexed. Some applications combine both, so the interface alone need not reveal the workflow.

A tall teal pitcher stands to the left of a shorter ivory pitcher.
This generated image meets the count, color and ordering constraints in the brief. It depicts an imagined still life.

How would you describe this result responsibly?

Record the brief and identify the illustration as generated. For a factual guide, verify names, geometry, labels and quantities against independent sources. A prompt states the desired result; inspecting the delivered image establishes which visual constraints it met.

Follow the idea

Diffusion Models · Learn one mechanism for synthesizing images.

Inpainting · See how generation can be limited to part of an image.

Source · Rombach and colleagues on latent diffusion models

Return to Image Generation in the A–Z glossary