DALL·E 2 introduced many of us to powerful AI image generation. Once OpenAI lifted the waitlist restriction, anyone could type a prompt into a text box and see their ideas brought to life by artificial intelligence. Now, OpenAI is soon bringing DALL·E 3 to the world, not only improving on DALL·E 2’s image generation, but also integrating directly with ChatGPT.
What’s new with DALL·E 3
OpenAI didn’t get into the weeds with its DALL·E 3 announcement, but we do know about some major enhancements over DALL·E 2. For starters, OpenAI claims DALL·E 3 does a much better job incorporating all aspects of any given prompt. Where DALL·E 2 might drop a concept or two in processing, DALL·E 3 supposedly produces content that takes all requests into consideration.
OpenAI also touts general improvements in DALL·E 3‘s outputs. Given the same prompt, DALL·E 3 should produce a more accurate, vivid, and impressive result over DALL·E 2.
But perhaps most notable of all, DALL·E 3 works within ChatGPT Plus. So long as you pay for the premium version of ChatGPT, you’ll be able to generate images directly in your chats. Plus, ChatGPT will expand on your prompts, suggesting more detailed requests that will allow DALL·E 3 to produce a better image. If you aren’t happy with the end result, ask ChatGPT to tweak it however you like.
OpenAI demonstrated the new ChatGPT/DALL·E integration to The Verge’s Emilia David. In one scenario, they asked ChatGPT to generate a logo for a ramen restaurant in the mountains. ChatGPT took it upon itself to come up with a longer prompt, then used that to generate four options with DALL·E 3. David’s favorite was a “rendering of a mountain with ramen snowcaps, broth flowing down like a waterfall, and pickled eggs on the ground like garden stones,” although they admitted the result looked more fit for a consumer product than something you’d expect for a restaurant.
DALL·E 3 has a focus on safety
OpenAI says DALL·E 3 is designed to decline generation requests for a public figure, at least by name. They also put time into preventing generation of biases towards visual over and under representation, as well as propaganda and misinformation.
The company is also looking into ways to note when an image was created with AI, starting with a “provenance classifier,” which will help them internally when determining when DALL·E 3 was used to generate an image.
In addition, OpenAI will let artists block their work from being used in future versions of these image generation models by submitting a piece they own the rights to. Future iterations of DALL·E will not produce results that resemble that blocked image and style. It’s a good development as the world reckons with the ethics of generating AI art.
When can I try DALL·E 3?
DALL·E 3 isn’t out quite yet, as it’s still in “research preview.” However, OpenAI says the service will be available on ChatGPT Plus and Enterprise customers beginning in October.
Leave a Reply
You must be logged in to post a comment.