OpenAI is opening up DALL-E 3 access
OpenAI is making its newest iteration of its text-to-image tool more widely available. On Thursday, OpenAI started letting ChatGPT Plus and Enterprise customers use the new DALL-E 3 model inside the ChatGPT app.
OpenAI says it has put safety measures in place to get DALL-E 3 ready for more people to use. DALL-E 3 was first announced last month, and OpenAI showed how it improves on the previous DALL-E 2. Users can now use ChatGPT to write longer and more descriptive prompts to feed into DALL-E 3, which helps it generate better images.
The generative image tool was also added to Bing Chat and Bing Image making Microsoft’s platform the first to give more public access to DALL-E 3 even before ChatGPT.
However, the promised safeguards to prevent harmful images haven't always worked. Users made images of cartoon characters crashing planes into the World Trade Center buildings. Even after Microsoft blocked certain prompts, simple workarounds produced similar results.
Text-to-image tools like Midjourney, Stable Diffusion, and older DALL-E versions have also caused controversy. The tech has copied copyrighted images, created fake nudes, changed the ethnicity of people, and falsely showed public figures.
OpenAI says it has taken way more steps this time to address these issues, and recently launched a website explaining the research for DALL-E 3 safety.
The company says it will "limit the chance of generating art in living artists' styles, images of public figures, and improve representation across images." OpenAI also has an internal tool that it says can detect DALL-E 3 images with 99 percent accuracy.