Multimodality · 2022
Latent Diffusion (Stable Diffusion) / Diffusion Transformers (DiT)
Introduced latent diffusion (the basis of Stable Diffusion), which runs the diffusion process in a compressed autoencoder latent space instead of pixel space, cutting the compute cost of training and sampling high-resolution image generators by a large factor.
Editorial record
Plain-language summary
An autoencoder first compresses images into a smaller latent representation, and the denoising diffusion model learns to generate in that latent space rather than over millions of raw pixels. A cross-attention mechanism injects conditioning such as text prompts, and the decoder turns the final latent back into a full-resolution image. Working in the smaller space made training and inference cheap enough to run on consumer GPUs, which enabled the open release of a general-purpose text-to-image model.
Knowledge graph
Relationships
Antecedents
Parallel developmentEvidence: Strongly supported
Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding (Imagen)
Imagen and Latent Diffusion are parallel text-to-image diffusion systems
P-532
Depends onEvidence: Strongly supported
Video Generation Models as World Simulators (Sora)
Sora is a diffusion Transformer applied to video
P-533
Descendants
Applies toEvidence: Direct
Vision Transformer (An Image is Worth 16x16 Words)
DiT uses a Transformer denoiser
P-308
EnablesEvidence: Strongly supported
LAION-5B: An Open Large-Scale Dataset for Training Next Generation Image-Text Models
LAION-5B provided the image-text data used to train Stable Diffusion
LAION
Source record
Provenance
- Record ID
- P-308
- Record created
- 2026-07-13
- Last reviewed
- 2026-07-14
- Record version
- 2
- https://arxiv.org/abs/2112.10752
- arXiv:2112.10752
Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.