Downcodes editor reported: OpenAI released the revolutionary AI image generation technology sCM, which has made breakthrough progress in image generation speed and efficiency and can be called a milestone in the field of AI images. By simplifying the calculation steps, the sCM model achieves the magical effect of generating high-quality images in just two steps, and is 50 times faster than the traditional diffusion model. It only takes 0.11 seconds to generate an image on the A100 GPU, and the maximum model parameters reach 1.5 billion, setting a new industry record. This technology not only achieved excellent test results on the CIFAR-10 and ImageNet data sets, but also showed strong expansion potential, laying a solid foundation for the training of larger-scale models in the future.
Technical core advantages:
Generate high-quality images in just two calculation steps
It only takes 0.11 seconds to generate an image on the A100GPU
Compared with the traditional diffusion model, the speed is increased by 50 times
The maximum model parameters reached 1.5 billion, setting a new record
In actual tests, sCM performed impressively. It achieved an FID score of 2.06 on the CIFAR-10 dataset and an excellent score of 1.88 when generating 512x512 pixel images on ImageNet. These metrics lag only about 10% behind the best existing diffusion models, but represent a qualitative leap in speed.
The key to technological innovation is to solve the fundamental problems of the traditional consistency model. Previous models used discrete time steps, which not only required additional parameters but were also error-prone. OpenAI's research team unified various methods by establishing a simplified theoretical framework and successfully identified and solved the main causes of training instability.
What is even more exciting is that this technology shows strong expansion potential. OpenAI successfully trained a model with 1.5 billion parameters on the ImageNet data set, which is the first time for a similar model. The study found that as the model size increases, the image quality continues to improve, which means that larger-scale model training may be possible in the future.
The advent of sCM technology heralds a new wave of development in the field of AI image generation. Its huge improvement in speed and efficiency will provide possibilities for more application scenarios. Let us wait and see what OpenAI will bring in the future. A surprising technological breakthrough!