Description

Technical and conceptual exploration of latent diffusion models. Students learn forward/reverse diffusion, CLIP text embeddings, U-Net noise prediction, and spatial conditioning via ControlNet, IP-Adapter, and LoRAs.

Key Topics

  • Diffusion Mechanics: From Gaussian noise to coherent imagery via iterative reverse denoising steps.
  • Latent Space Geometry: Vector math in compressed latent dimensions and spherical linear interpolation (slerp).
  • Precision Control: Depth maps, Canny edges, OpenPose skeletons, and semantic segmentation guidance.

References & Wiki Links