Examples

1. 1.

   Stable Diffusion is a latent diffusion model - it encodes images to 64x64 latent representations via a VAE, runs diffusion in that space, then decodes back to 512x512 pixels, using 8x less memory than pixel-space diffusion.

2. 2.

   Runway Gen-2 uses latent diffusion conditioned on text, images, or video clips to generate video footage - used by film directors for previsualization and by advertising agencies for rapid concept prototyping.

3. 3.

   Adobe Firefly is built on latent diffusion models trained exclusively on licensed Adobe Stock content - providing enterprise users with copyright-safe image generation integrated directly into Photoshop and Illustrator.