2.1.24

This AI imagery tool can transform famous paintings into different styles



GIVE MONA LISA A MOHAWK WITH DALL·E 2

A couple of weeks ago, we reported on Google’s AI tool that can turn any text into a photorealistic image. Well, it turns out Google isn’t the only tech company vying for a slice of the AI image generator pie. Meet OpenAI, a San Francisco-based company that created its first text-to-image system back in January 2021. Now, the team has unveiled its latest system, called ‘DALL·E 2’, which generates more realistic and accurate images with 4x greater resolution.

Both Imagen and DALL·E 2 are tools that use artificial intelligence to transform simple text prompts into photorealistic images that have never existed before. As explained in the video above, DALL·E 2 can also make realistic edits to existing images, meaning you can give famous paintings different styles or even give Mona Lisa a mohawk. The AI system was created by training a neural network on images and their text descriptions. Through deep learning, DALL·E 2 can identify individual objects and understand the relationships between them. OpenAI explains, ‘DALL·E 2 has learned the relationship between images and the text used to describe them. It uses a process called ‘diffusion’, which starts with a pattern of random dots and gradually alters that pattern towards an image when it recognizes specific aspects of that image.

OpenAI says its mission is to ensure that artificial intelligence benefits all of humanity. The company says, ‘Our hope is that DALL·E 2 will empower people to express themselves creatively. DALL·E 2 also helps us understand how advanced AI systems see and understand our world, which is critical to our mission of creating AI that benefits humanity.’

However, despite the company’s intentions, this kind of technology is a tricky one to deploy responsibly. With this in mind, OpenAI says it is currently studying the system’s limitations and capabilities with a select group of users. The company has already removed explicit content from the training data to avoid violent, hate, or adult images being generated. They also say that DALL·E 2 cannot generate photorealistic AI versions of real individuals’ faces.

design: OpenAI