Openai is expanding access to its latest text-to-image generator, DALL-E 3. Starting Thursday, ChatGPT Plus and Enterprise subscribers will be able to utilize the new model within the ChatGPT app. OpenAI has implemented a safety mitigation stack to ensure a secure release of the model.
Last month, DALL-E 3 was introduced as an enhancement over the previous DALL-E 2. It enables users to create longer and more visually descriptive prompts in ChatGPT, which in turn generates corresponding images. Prior to being made available in ChatGPT, DALL-E 3 was already incorporated into Bing Chat and Bing Image Generator, making it the first text-to-image model to be accessible to the general public through Microsoft's platform.
However, there have been concerns about harmful output from text-to-image generators in the past. For instance, users have created images depicting copyrighted materials, nonconsensual images, altered ethnicity of subjects, and photo-realistic misrepresentations of public figures. OpenAI claims to have taken significant measures to address these issues with DALL-E 3. They have released a website showcasing their research on the model and assure users that it has been trained to limit the generation of content resembling the style of living artists and images of public figures, while also improving demographic representation in the generated images. OpenAI has also implemented an internal "provenance classifier" tool to detect with 99 percent accuracy if an image was generated by DALL-E 3.