DALL-E 3, the latest version of OpenAI’s AI image generator, is now live for premium users.
ChatGPT Plus and Enterprise users now have access to the company’s advanced image model. With DALL-E 3 directly integrated into the chat interface, users can generate photorealistic images from text prompts. The company claims in the announcement that DALL-E 3 can render hands, text, and faces — something that AI photo generators have typically had problems with doing accurately.
SEE ALSO:
OpenAI is working on a tool to detect DALL-E 3 AI-generated images
Tweet may have been deleted
OpenAI also shared how it’s safeguarding against potential nefarious uses of its image-generating software. The model has a “multi-tiered safety system” that limits DALL-E 3’s ability to generate violent, hateful, or adult content. This includes passing user prompts through a safety check before responding with the generated image. The model was stress-tested by humans, who red-teamed it to look for faults — with OpenAI publishing the results of these tests in a research paper. The announcement also shared a tool OpenAI is developing that can detect DALL-E 3 generated images, called a “provenance classifier.”
Addressing concerns of potential copyright and plagiarism violations, OpenAI says it is limiting the model’s “likelihood of generating content in the style of living artists, images of public figures, and to improve demographic representation across generated images.” Additionally, OpenAI offers an option for artists and creators to block their works from being used to train the model which, as some users pointed out, is hilariously tedious, since it requires listing every single work to be excluded in an online submission form.
Tweet may have been deleted
If all that sounds a bit vague to you, you’re not alone. We have a lot of questions. How “limited” is the model? What about content in the style of dead artists that are still copyrighted? What about generating imagery related to non-public figures? Mashable has reached out to OpenAI for clarification on its safety and responsibility measures and will update this story if it responds. Since the tool is live today for Plus and Enterprise users, we’re sure to find out soon just how safeguarded, jailbreak-able, and copyright-protected it really is.