OpenAI Introduces New Consistency Model With Enhanced Training Stability

According to BlockBeats, on October 24, OpenAI announced the launch of its latest consistency model, which features simplified formulas, higher training stability, and scalability. The new sCMs (simplified Consistency Models) are capable of generating samples comparable to leading diffusion models but require only two sampling steps. The consistency model offers a faster alternative for generating high-quality samples compared to traditional diffusion models. Unlike diffusion models that gradually generate samples through numerous denoising steps, the consistency model is designed to convert noise directly into noise-free samples in a single step.