Latent Diffusion
Latent Diffusion is an image generation architectural approach in which the diffusion process is performed in a low-dimensional latent space produced by an encoder, rather than in the high-dimensional pixel space, with a decoder subsequently reconstructing the image. Compared to pixel-space methods, this approach significantly reduces computational resource requirements while maintaining high generation quality, and forms a foundational technique for modern text-to-image models such as Stable Diffusion.
Core Facts
Latent diffusion models significantly reduce computational costs by performing the diffusion process in a low-dimensional latent space rather than pixel space.
The latent diffusion architecture uses a Variational Autoencoder (VAE) to compress images into a low-dimensional latent space, performing noise addition and denoising in the compressed space, which is the origin of the name Stable Diffusion.
Timeline (last 90 days)
The latent diffusion architecture uses a Variational Autoencoder (VAE) to compress images into a low-dimensional latent space, performing noise addition and denoising in the compressed space, which is the origin of the name Stable Diffusion.
The latent diffusion architecture reduces computational load by tens of times, making it possible to run large-scale image generation on consumer-grade GPUs.
Latent diffusion models significantly reduce computational costs by performing the diffusion process in a low-dimensional latent space rather than pixel space.
All Facts (3)
Latent diffusion models significantly reduce computational costs by performing the diffusion process in a low-dimensional latent space rather than pixel space.
85%VerifiedThe latent diffusion architecture uses a Variational Autoencoder (VAE) to compress images into a low-dimensional latent space, performing noise addition and denoising in the compressed space, which is the origin of the name Stable Diffusion.
65%UnverifiedThe latent diffusion architecture reduces computational load by tens of times, making it possible to run large-scale image generation on consumer-grade GPUs.
50%