Artificial Intelligence Producing Images from Text Will Be 30 Times Faster

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

Currently available text-to-image generation tools generally produce very good results in one go. they don’t give. The new framework developed by MIT engineers addresses exactly this point. Framework streamlines the image generation process in a single step. to 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 the machine learning techniques.teacher StudentIt is based on “. In this technique, models that have already been trained are transformed into a new model by imitating the original ones. Tianwei Yin from MIT states that with this method, both the visual quality is increased and the existing model is improved. up to 30 times He stated that it was accelerated. Thus, the user will not need 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 to all artificial intelligence tools They believe it can be adapted. 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’s not easy let’s say. Because MIT introduced DMD technology using pre-trained networks. In other words, already existing technologies were reworked. In this way, the process was accelerated. Desire to use DMD in more comprehensive projects, much larger training sets are needed will lead.