麻省理工大学（MIT）与Adobe公司近日合作研发了一种名为DMD的技术，该技术能够在不影响图像质量的情况下，显著提升图像生成速度。DMD全称为Distribution Matching Distillation，它能够将多步扩散模型简化为一步图像生成解决方案。据研究人员称，DMD模型可以在现代GPU硬件上每秒生成20幅图像，这一速度远超目前主流文生图模型的渲染时间。这项技术的出现，不仅能够提高图像生成的效率，还能够为相关领域的发展带来新的可能。
Title: MIT and Adobe Collaborate on AI Technology to Accelerate Image Generation
Keywords: MIT, AI Image, Generation Speed
News content: MIT, in collaboration with Adobe, has recently developed a technology named DMD that can significantly enhance the speed of image generation without compromising image quality. DMD, which stands for Distribution Matching Distillation, simplifies the multi-step diffusion models into a one-step image generation solution. Researchers claim that the DMD model can generate 20 images per second on modern GPU hardware, a speed far exceeding that of the current mainstream text-to-image models. The emergence of this technology not only increases the efficiency of image generation but also opens up new possibilities for the development of related fields.
【来源】https://www.ithome.com/0/758/402.htm
Views: 1