OpenAI opens access to DALL-E 3

OpenAI is launching its latest text-to-image generator, making it more versatile. The company will provide access to the new DALL-E 3 model within the ChatGPT application to ChatGPT Plus and Enterprise customers starting on Thursday. OpenAI has prepared a secure mitigation stack for the model, enabling it to be ready for scalable deployment. DALL-E 3 was first released last month, showcasing improvements over the previous DALL-E 2 by allowing users to provide longer and more visually descriptive prompts to the image generator through ChatGPT. DALL-E 3 has been added to Bing Chat and Bing Image Generator, making Microsoft the first platform to introduce the model, even before ChatGPT. The advertised barriers to reduce harmful images are not always effective, as other simple methods fail to achieve desired results even after Microsoft has blocked certain prompts. Text-to-image generators like Midjourney, Stable Diffusion, and earlier iterations of DALL-E have faced controversies of their own. These technologies have produced copyrighted image materials, non-consensual nude photos, racial variations of objects, and realistic distortions of public figures. OpenAI promises to take more effective measures this time and has provided a website showcasing the research behind DALL-E 3. The company states that it will "constrain the model's generation of content in the style of specific artists and public figure images, while improving the representativeness of generated images." OpenAI also has an internal "source classifier" tool that claims to detect whether an image is generated by DALL-E 3 with 99% accuracy.