OpenAI announces Dall-E 3, will be integrated into ChatGPT
OpenAI has announced Dall-E 3. The third iteration of the image generator will be integrated into ChatGPT. That tool can be used to improve prompts. According to OpenAI, Dall-E 3 creates better images and has more safeguards against algorithmic biases.
OpenAI announced the tool, but it won’t be released until October. Then Dall-E 3 can be used by subscribers with a ChatGPT Plus or Enterprise package. The tool is currently available in beta. The tool is particularly distinguished by its integration into ChatGPT. Users will soon be able to provide a prompt there to have an image created immediately. Users can also submit improvements or changes to the image via chat if they want to make adjustments.
According to OpenAI, this should ensure that users can generate a better image with less text. Currently the rule of thumb with Dall-E 2 or other image generators is that the image becomes better or sharper if the prompt is more verbose. Thanks to the integration with ChatGPT, this is no longer necessary. “Modern text-to-image generators tend to ignore words or descriptions,” OpenAI writes. “Users are therefore forced to learn engineering promptly.” According to the makers, this is no longer necessary with Dall-E 3.
The makers of the generator have also paid more attention to safety on the platform in version 3. This makes it more difficult to create violent, hateful or ‘adult’ images. OpenAI has deployed a red team for improvements in ‘risk areas’. Biases have also been included in the algorithm; it concerns ‘biases related to visual over- or under-representation’.
It will also no longer be possible to provide prompts around public figures. OpenAI hopes to combat propaganda and disinformation. OpenAI is also testing an internal tool that should recognize when an image was created by Dall-E 3. “We hope to use this tool to understand how generated images are created,” the company says.
OpenAI also says it is taking measures to help artists protect their own intellectual property. Dall-E 3 rejects prompts that create images in the style of certain artists. Artists can also ensure that their art is not used by AI for training purposes. However, that process is cumbersome and not very extensive. It concerns an opt-out where artists must submit images together with a form. OpenAI will then ‘study’ this and only then can it decide whether it will no longer include the images for future training models.