From the source
From the source
Runway researchers introduced latent diffusion models (LDMs) that apply diffusion in the latent space of pretrained autoencoders, using cross-attention layers for conditioning inputs such as text or bounding boxes.
LDMs achieve competitive performance on unconditional image generation, inpainting, and super-resolution while reducing computational requirements compared to pixel-based diffusion models.
From the source
By decomposing the image formation process into a sequential application of denoising autoencoders, diffusion models (DMs) achieve state-of-the-art synthesis results on image data and beyond. ... Our latent diffusion models (LDMs) achieve highly competitive performance on various tasks, including unconditional image generation, inpainting, and super-resolution, while significantly reducing computational requirements compared to pixel-based DMs.
runway.com
