Diffusion

Diffusion model projects from denoising fundamentals to latent generation and sampling efficiency.

DDPM from Scratch Forward noising, denoising objective, beta schedules, and image generation loop

U-Net Denoiser Residual blocks, time embeddings, attention blocks, and multi-scale feature maps

Latent Diffusion Autoencoder latents, conditioning, classifier-free guidance, and image synthesis

Fast Sampling DDIM-style sampling, scheduler comparisons, and quality-latency tradeoffs