Massachusetts Institute of Technology (MYTHA group of scientists working within ) have developed a new technology that makes artificial intelligence tools such as DALL-E 3 and Stable Diffusion, which are used to produce images from text, up to 30 times faster. frame They announced their development. This framework will enable users to achieve the best results in the shortest time possible.

The currently available text to image generation tools generally produce very good results the first time. they don’t care. The new framework developed by MIT engineers addresses exactly this point. Framework streamlines the image generation process into one step. simplify and focuses on faster image production. Moreover, these images are in HD quality.

So how was this possible?

The framework, which MIT engineers call “DMD,” is one of several machine learning techniques.Student teacherIt is based on “. This technique converts already trained models into a new model by imitating the original one. Tianwei Yin of MIT states that this method both increases visual quality and improves the existing model. up to 30 times He stated that it was accelerated. The user does not have to process input again and again.

DMD can go far beyond producing images

MIT designed DMD to generate images from texts and achieved successful results. However, this does not mean that technology will only be used to produce images from text. MIT engineers say this for all artificial intelligence tools They believe it can be customized. If such a thing were to become a reality, DMD, 3D modeling or… pharmaceutical production It can also be used in areas where high speed is more important, such as

However, this is not enough It is not easy let’s say. Because MIT introduced DMD technology using pre-trained networks. In other words, pre-existing technologies were reworked. In this way the process was accelerated. Desire to use DMD in more comprehensive projects, Much larger training sets are required will suffer.