Similar to DALL-E 2, the artificial intelligence system developed by the US-based OpenAI Google Imagen after that Party is also on the agenda.
Artificial intelligence systems are changing many things, and these now include the fields of art / design. Trained with millions of different photos, these systems now make designing something child’s play. It is certainly the most resonant of these recently. DALL-E 2 it happened. Thanks to this technology, it can create realistic designs / drawings / photographs with simple sentences written. introduced last month Google Imagen It offers exactly that. The system, which can create impressive photos / designs from the entered sentences, can provide similar results with DALL-E 2 and even better in many examples. However, as it turns out Google just that’s not enough. One in the process Party work on being done.
imagen with Party Although the focus is the same, different foundations are used. According to Google’s statement imagen on the side “diffusion‘ model is used. This model can learn to transform a pattern of random dots into images. Google here”These images start at low resolution first and then gradually increase the resolution.” says.
Party’s approach to the process is based on the conversion of a special collection of images into a series of code entries, similar to puzzle pieces. The sentence / text input entered in the process is converted to these code entries and a new image is created. According to the description “autoregressive” Working from the logic of the Party, it can handle longer sentences and produce more realistic results. These systems, which have a very complex technical basis and therefore are not overly detailed, impress everyone and promise a great future.
YOU MAY BE INTERESTED
DALL-E 2 an influencer that establishes a link between writing and creativity. Google Imagen and Party systems are not yet publicly available. Behind this is the possibility of abuse. Like DALL-E 2, Google wants to rid their systems of dangerous/fake use cases before making them public.
By the way, the graphics card giant Nvidia it also works. Developed by the company in a similar structure GauGAN2 system introductory video directly here You can watch.
-This example by Google shows the result of different approaches. The result of different approaches can be understood in the image at the top of the page.