OpenAI, a research company that aims to create artificial intelligence that can benefit humanity, has announced a new feature for its chatbot ChatGPT: the ability to create images from text using its latest model, DALL-E 3.
What is DALL-E 3 and how does it work?
DALL-E 3 is an AI system that can generate images from detailed textual prompts provided by users. It leverages ChatGPT’s natural language processing capabilities to understand the context and nuances of the text, and then produces images that match the description.
DALL-E 3 is an improvement over its predecessor, DALL-E 2, which was released in January 2021. According to OpenAI, DALL-E 3 can generate images that are not only more visually striking but also crisper in detail. It can also handle longer and more complex prompts, and support both landscape and portrait aspect ratios.
DALL-E 3 is built on ChatGPT, which is a chatbot that can have natural conversations with users on various topics. ChatGPT can also act as a brainstorming partner and a refiner of the prompts for DALL-E 3. Users can ask ChatGPT what they want to see in anything from a simple sentence to a detailed paragraph, and ChatGPT will automatically generate tailored prompts for DALL-E 3 that bring the idea to life.
How can users access DALL-E 3?
DALL-E 3 is now available to ChatGPT Plus and Enterprise customers through the ChatGPT app. Users can simply describe their vision in the chat, and ChatGPT will provide a selection of visuals for them to choose from. Users can also ask for revisions or tweaks right in the chat.
DALL-E 3 is also integrated with Microsoft’s Bing Search and Bing Chat, which became the first major platforms to deploy DALL-E 3 publicly last month. Users can use Bing Search to find images based on text queries, or use Bing Chat to create images from text conversations.
As with DALL-E 2, the images created with DALL-E 3 are owned by the users and they do not need OpenAI’s permission to use them for any purpose.
What are the safety measures for DALL-E 3?
OpenAI has taken steps to limit DALL-E 3’s ability to generate potentially harmful imagery, such as violent, adult, or hateful content. It has implemented a multi-tiered safety system that runs checks over user prompts and the resulting imagery before it is surfaced to users.
OpenAI has also worked with early users and expert red-teamers to identify and address gaps in coverage for its safety system, which emerged with new model capabilities. For example, it has improved its performance in areas such as generation of public figures, graphic content, and harmful biases.
Additionally, OpenAI has taken measures to limit DALL-E 3’s likelihood of generating content in the style of living artists, images of public figures, or misleading images. It has also developed an internal tool called a provenance classifier that can identify if an image was generated by DALL-E 3 with over 99% accuracy.
OpenAI says it will continue to hone DALL-E 3’s safety through user feedback and expert guidance, and share more updates on its research and development.
