- DALL-E 3 is now integrated with ChatGPT, making it easier than ever to create art.
- OpenAI has gone all-in on safety features to keep things clean and respectful.
- Artists now have the option to opt out of having their artwork used to train future OpenAI text-to-image models.
September 21, 2023: OpenAI has introduced the third iteration of its text-to-image AI tool, DALL-E 3. This new version brings with it a unique integration with ChatGPT.
Now, users don’t have to think too much about creating their prompts.
Instead, they can just ask ChatGPT to generate a prompt for them. This change makes it easier for people to get into AI art, according to OpenAI.
Lead researcher and DALL-E team head, Aditya Ramesh, showcased how the feature works. He asked ChatGPT to help create a logo for a ramen restaurant in the mountains.
Really proud of my team @neonbjb @gabeeegoooh@jingli9111 for shipping DALL·E 3 in ChatGPT and starting to get it in peoples’ hands! It was a privilege working with them over the past year. It’s been fun seeing how things have progressed since the first avo chair and I’m excited… pic.twitter.com/5cRuDFeeFX
— Aditya Ramesh (@model_mechanic) September 20, 2023
ChatGPT generated a detailed prompt, and DALL-E 3 produced several options, one of which depicted a mountain with ramen snowcaps.
OpenAI pointed out that this makes art creation more accessible to people who might struggle with generating their own prompts.
OpenAI first launched DALL-E in January 2021. By the time DALL-E 2 came around in 2022, the company had to put a waitlist due to the platform’s capability to generate explicit and biased images.
This waitlist was removed last September, and DALL-E 2 was opened publicly. The new version, DALL-E 3, is set to roll out first to premium ChatGPT users in October and then to research labs and API customers in the fall.
OpenAI has not confirmed when a free public version will be available.
For this release, OpenAI has emphasized safety. Sandhini Agarwal, a policy researcher at OpenAI, stated she has “high confidence” in the new safety measures. These include the use of external red teamers and input classifiers to prevent the creation of inappropriate or harmful images.
DALL-E 3 will also decline to mimic the style of living artists and cannot generate images of public figures if their names are specifically mentioned in the prompt.
Artists can also protect their work now. OpenAI will allow them to submit a request form to remove their art from future text-to-image AI models.
This move comes after artists sued competitors Stability AI and Midjourney for using their copyrighted work.
The launch of DALL-E 3 is part of the growing competition in generative AI, especially in the realm of image creation. Other platforms like Stability AI and Midjourney are continuously refining their models, putting pressure on OpenAI to keep pace.
