Incredible AI system DALL-E 2 enters public beta
Last month, we shared unique AI-generated camera designs inspired by pop culture icons like Darth Vader and Batman. The camera ideas were developed by photographer Mathieu Stern, but the designs themselves were created by DALL-E 2, an AI system built by OpenAI. At that time, DALL-E 2 was limited to a private beta test, so readers wanting to use AI to create cameras of their own were out of luck. However, OpenAI has announced that DALL-E is now available as a public beta. The company will invite a million people from its waitlist in the coming weeks.
If you’re invited into the public beta test, you’ll be able to create with DALL-E using free credits that refill each month. During the first month, users will receive 50 free credits. Each subsequent month, the free credit allocation drops to 15. Each DALL-E prompt generation requires one credit and delivers four images. You can also edit or create a variation prompt, which returns three additional images. If those aren’t enough credits, OpenAI sells additional credits in 115-generation increments for $15. It’s worth noting that the number of images generated by each prompt is ‘approximate.’
In case you aren’t familiar with DALL-E, it’s an AI system that generates realistic images and art from a natural language text description. For example, if you typed ‘a painting of a fox sitting in a field at sunrise in the style of Claude Monet,’ that’s what DALL-E 2 would generate.
|‘A painting of a fox sitting in a field at sunrise in the style of Claude Monet.’ DALL-E 1 (left) vs DALL-E 2 (right). DALL-E generates more realistic, accurate results with four times the resolution of the original DALL-E.|
It’s the successor to the original DALL-E that OpenAI released in 2021. Compared to the original iteration, DALL-E 2 promises more realism and accuracy, plus four times greater resolution. According to research, nearly 72% of respondents preferred DALL-E 2’s caption matching and about 89% preferred it for its photorealism.
DALL-E 2 works because it has learned how descriptive text relates to images. The system uses a process called ‘diffusion.’ With diffusion, DALL-E starts with a pattern of random dots and gradually changes the pattern of dots toward an image ‘when it recognizes specific aspects of that image.’
Once DALL-E 2 has created an image from a text prompt, you can edit the results. You can change the location of requested objects, add or remove elements, and adjust shadows, reflections, colors, textures, and more.
|DALL-E 2 allows users to edit the results. For example, what if you wanted to add a corgi to the scene on the left? You can describe where you want it located, such as in the painting of a man (right).|
When DALL-E 2 entered private beta, one reason not just anyone could use the AI was concerns about harmful content. OpenAI has been developing safety mitigations. DALL-E 2’s training data has been cleaned up to remove the most explicit content. The team has also developed techniques to prevent the photorealistic generation of real people, including public figures. DALL-E 2 has a content filter to identify text prompts and image uploads that violate OpenAI’s policies. Plus, there are automated and human monitoring systems in place.
There’s no doubt you could have a lot of fun with DALL-E 2. Whether you want a children’s book illustration of astronauts playing basketball in space with cats, teddy bear scientists mixing chemicals, or something else entirely, DALL-E 2 can generate it. If you’d like to try it for yourself, visit OpenAI and sign up for the waitlist.