Planned chapter | Dan Hollick
Generating images.
Modern image models learn to reverse a corruption process, gradually turning noise into a structured image guided by text or another picture.
The source chapter is still planned. This route preserves the collection and offers an original conceptual preview.
Modern image models learn to reverse a corruption process, gradually turning noise into a structured image guided by text or another picture. The apparent simplicity comes from a set of carefully chosen representations, transformations and physical assumptions working together.
Diffusion
Training teaches a network to predict and remove noise at many levels of corruption.
This is one part of a longer chain: noise becomes condition becomes denoise steps becomes image. The useful abstraction hides the physical work, but the underlying constraints still shape the software built above it.
Latent space
Compression allows denoising to happen in a smaller representation than full-resolution pixels.
The implementation is full of compromises. Precision, speed, storage and energy rarely improve together, so practical systems choose the errors people are least likely to notice.
Guidance
Text embeddings steer each denoising step toward requested visual concepts.
Once this layer is visible, familiar design conventions stop looking arbitrary. They are accumulated responses to the capabilities and limits of the machinery below.
A visual study based on the original chapter. Text is condensed and rewritten.