dalle1-not-diffusion-uses-autoregressive-vae

IN premise — summaries/2026/08/24/wiki-Transformer_deep_learning_architecture-chunk-4-chunk-2.md

Created 2026-08-24T17:11:25+00:00

DALL-E 1 is not a diffusion model; it uses a decoder-only autoregressive Transformer for image token generation plus a VAE decoder for pixel reconstruction, distinguishing it from later diffusion-based models (Stable Diffusion, DALL-E 2/3)

Summary

DALL-E 1 builds images the way a text autocompleter builds sentences, predicting one visual token after another and then reconstructing pixels from the final sequence, rather than starting with random noise and gradually refining a picture the way diffusion models do. This matters because DALL-E 1 belongs to a fundamentally different architectural family than Stable Diffusion or DALL-E 2/3, so assumptions about how one class of model works do not transfer to the other.