OpenAI releases third version of DALL-E

OpenAI has announced the release of the third version of its generative AI visual art platform, DALL-E. This updated version now includes integration with ChatGPT, allowing users to use the chatbot to create prompts for DALL-E. Additionally, DALL-E 3 comes with enhanced safety options to prevent the generation of explicit or potentially hateful images.

DALL-E, which was first launched in January 2021, converts text prompts into images. However, earlier versions often struggled with specific wording and failed to understand context effectively. OpenAI researchers claim that the latest version has made significant improvements in understanding context and generating more accurate and relevant images.

The integration with ChatGPT is one of the notable features of DALL-E 3. Instead of users having to come up with their own detailed prompts, they can now rely on the chatbot to create prompts for DALL-E. ChatGPT will generate a paragraph, and DALL-E 3 will use it as a guide to produce the desired image. This integration broadens the appeal of DALL-E to a wider audience, as users no longer need to be proficient in creating prompts themselves.

In a recent demonstration to The Verge, Aditya Ramesh, the lead researcher and head of the DALL-E team, used ChatGPT to come up with a logo for a ramen restaurant situated in the mountains. DALL-E generated several options based on the prompt, including a rendering of a mountain with ramen snowcaps, broth flowing down like a waterfall, and pickled eggs on the ground like garden stones. This integration with the chatbot simplifies the image creation process and allows more people to engage in AI art generation.

OpenAI has put considerable effort into enhancing the safety measures of DALL-E 3. The company worked with external red teamers, who intentionally attempt to break a system to test its safety, to ensure that explicit or potentially hateful images are not generated. Input classifiers were leveraged to train the language model to ignore certain words and prevent explicit or violent prompts. Additionally, DALL-E 3 is designed not to recreate images of public figures unless specifically mentioned in the prompt.

The release of DALL-E 3 will be staggered, with ChatGPT Plus and ChatGPT Enterprise users getting access first in October. Research labs and the API service will follow in the fall. OpenAI has not yet provided a specific timeline for when a free public version will be released.

To address concerns regarding copyright infringement, OpenAI has allowed artists to request the removal of their art from future versions of text-to-image AI models. Artists can submit an image that they own the rights to through a form on OpenAI’s website. A future version of DALL-E will then block results that resemble the artist’s image and style. This move is likely meant to prevent potential lawsuits, as artists have sued similar platforms in the past for allegedly using their copyrighted work to train their models.

OpenAI representatives have expressed confidence in the safety measures implemented in DALL-E 3. However, they acknowledge that the model is not perfect and will continue to be refined and improved over time.

Overall, the release of DALL-E 3 with integrated ChatGPT and enhanced safety measures marks a significant advancement in AI-generated visual art. With its improved understanding of context and simplified prompt creation, DALL-E aims to make AI art creation more accessible and user-friendly.