How does Stable Diffusion work?
- User
How does Stable Diffusion work?
- Assistant
Stable Diffusion works by repeatedly applying a denoising autoencoder guided with text input.
An autoencoder tries to compress an image into a small vector then decompress back to the original image. Textual guidance can be added to improve its efficiency, by specifying what content it should or should not generate. A denoising autoencoder has the primary aim of restoring an image combined with noise back to the form before adding noise.
When we apply such autoencoder repeatedly to any image, the output will become less noisy and more resembles the input text. With enough iterations, we can get a clean image even when we feed pure noise as input. This workflow of repeatedly applying an autoencoder, together with a properly trained autoencoder on wide variety of objects, forms a Stable Diffusion model.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.