Ziv Ilan

This talk explores techniques to accelerate diffusion models for image and video generation, addressing the challenge of high latency in practical applications. The core thesis is that by employing methods like quantization, caching, and distillation, developers can significantly reduce the number of diffusion steps required, thereby improving performance and enabling real-time generation without sacrificing quality.