OpenAI is unveiling a new version of its text-to-image prompting tool, DALL-E. DALL-E 3 can use ChatGPT to help with prompt ideas. Here’s the latest.
Text-to-image prompts are how most generative AI models create the images you see. The image you see above was created using the latest version of DALL-E 3 using the following text prompt, “A 2D animation of a folk music band composed of anthropomorphic autumn leaves, each playing traditional bluegrass instruments, amidst a rustic forest setting dappled with the soft light of a harvest moon.”
DALL-E3 uses ChatGPT to help it fill in prompts, available to subscribers of ChatGPT Premium plans. The direction the image takes can then be honed through conversations with the chatbot, with the results appearing directly in the chat app. That feature works similarly to Midjourney—which operates exclusively through Discord.
OpenAI says this new version of DALL-E can create higher-quality images that more accurately reflect the prompts fed to it. It also better handles generating content that has traditionally tripped up AI—including the generation of readable text and human hands that make anatomical sense.
One difference between this version of DALL-E and Midjourney? OpenAI says the chatbot will refuse to create art in the style of a living artist—or portray public figures. Artists can now opt out of having all of their artwork used to train future generations of artificial intelligence.
Though its unclear what OpenAI plans to do about those artists whose work was used to train the current version of the AI without permission. OpenAI is currently facing a major lawsuit from authors for training ChatGPT on their copywritten texts without their permission.
17 authors and The Authors Guild launched the lawsuit. Much like the Sarah Silverman lawsuit, OpenAI refuses to disclose or publicize the datasets used to train its AI models.