DALLE 2, an update to OpenAI’s text-to-image generator that appears to be a significant step forward, has been announced by OpenAI, the AI research startup.
DALLE 2 can, in essence, produce art from natural language input, such as “a painting of a fox resting in a field at daybreak in the style of Claude Monet.” The goal of OpenAI is to create “unique, realistic visuals and art” that can “combine concepts, qualities, and styles,” according to the company.
It’s a fantastic achievement for a business that has been working on a variety of AI projects, including GPT-3, which can write some really amazing text.
The research article that goes along with it from OpenAI goes into a lot more depth on how DALLE 2 works. DALLE 2 employs the CLIP contrastive models in two stages: the first makes a CLIP image embedded with a text caption, and the second generates an image from it.
The results are stunning, and they could have far-reaching implications for the art and graphic design sector, particularly for video game studios that now hire designers to meticulously create worlds and concept designs.
DALLE 2 expands on the work that OpenAI made with DALLE 1, which was unveiled in January 2021. Salvador Dal and WALLE are combined in the name.
DALLE 2 is a fantastic illustration of this technology in action, pushing the limits of what is possible when it comes to employing AI to generate visuals. Even existing photographs can be edited to a user’s specifications using the service.
Of course, there is still a long way to go. CLIP foundations can be readily deceived by labeling one object as another, according to the study. However, these issues will be ironed out and improved over time.
DALLE 2 will be released to a restricted group of academics by OpenAI in the near future.
