This presentation discusses DPM (diffusion probabilistic model) improvements. Key advancements include: (1) **Speedup** via skip-spacing, drastically reducing computation time (up to 80x faster). (2) **Deterministic sampling**, enabling consistent image generation from the same noise input, achieved by extending the diffusion process. (3) **Conditional generation**, allowing generation based on class information, text descriptions, or reference images (using techniques like classifier-free guidance, SD Edit, and LVR). These improvements enhance DPM's efficiency and versatility.