Dall-E is a tool that converts prompts from text to images. With the third version just released, OpenAI researchers say the tool understands context better, not missing specific words like the previous version.
Notably, Dall-E 3 is integrated with ChatGPT. Now users don’t have to come up with detailed reminders for Dall-E 3, they can just ask the ChatGPT chatbot to give them reminders. Users can also use their own reminders if they have specific ideas for Dall-E.
 |
| Image for Dall-E 3 generated. |
Dall-E was first released by OpenAI in January 2021, ahead of other platforms like Midjourney. With version two in 2022, OpenAI opened a waitlist to control who could use the tool after receiving criticism that Dall-E produced biased images. In September of that year, the US company removed the waitlist and opened Dall-E 2 to the public.
Dall-E 3 will be released to ChatGPT Plus and ChatGPT Enterprise users first in October, followed by labs and API services. A free version for everyone has not yet been announced.
OpenAI says it has put a lot of work into Dall-E 3 to create robust safety measures to prevent the creation of hateful images. The company has also worked with external “red actors”—groups that intentionally break a system to test its safety—and relied on input classification (a way to teach a language model to ignore certain words to avoid violent, provocative prompts). Dall-E 3 will also not be able to generate images of public figures if the prompt mentions a specific name.
In an email, an OpenAI representative said Dall-E 3 was trained not to create images in the style of living artists, unlike Dall-E 2.
Source
Comment (0)