OpenAI says DALL-E 3 is designed to turn detailed descriptions into images that more closely match what users ask for. Its planned integration with ChatGPT would let people describe an idea, review an image, and request changes in ordinary conversation.
More detail in the image brief
According to OpenAI, DALL-E 3 has improved at following complicated prompts. The company says it can better represent specific objects, how those objects relate to one another, and details such as human hands or text in an image—areas that have often challenged text-to-image systems.
The system is intended to support a broad range of visual approaches, including artistic abstracts, comic book drawings, and photorealism. As with any image generator, the written description remains central: it tells the model what should appear and how the parts of a scene fit together.
OpenAI presents the improvements as a way to make complex requests easier to express. Rather than relying on a carefully engineered string of instructions, users could explain the scene they want in more natural language.
ChatGPT can help shape and revise requests
DALL-E 3 is planned as a ChatGPT app for ChatGPT Plus users. In that setup, ChatGPT can act as a brainstorming partner, helping a user develop an idea and refine the description sent to the image system.
The conversation can continue after the first image appears. If the result is close but misses something, the user can tell ChatGPT what to change. OpenAI says this back-and-forth approach should reduce the need for elaborate prompt engineering.
That changes the interaction from writing one detailed instruction and accepting the result to an iterative process. A user can start with a broad concept, see what the system produces, and then clarify the parts that need adjustment.
Rollout and image ownership
OpenAI says images made with DALL-E 3 are fully owned by their creators, who can reuse, market, and sell them. The planned launch through ChatGPT Plus, ChatGPT Enterprise, and the API is in October. A separate version through OpenAI Labs, similar to DALL-E 2, is expected later this fall.
These are planned release details described by OpenAI. The article does not provide a more specific date for access or explain how availability might vary across the listed products.
Limits and questions around training
DALL-E 3 is described as rejecting prompts that name living artists, a restriction on a common way of asking image systems to follow a particular style. Prompts naming famous people are also expected to be rejected for safety reasons, including to reduce the risk of AI-generated images of politicians spreading on social media.
Artists will be able to exclude their images from training future OpenAI image models. Because the option applies to future models, it leaves open a question about whether artists' work may have been used to train DALL-E 3. The source says this could lead to further copyright discussions, and that OpenAI may be able to provide updated versions of DALL-E 3 after an opt-out period.
OpenAI says it plans to address risks such as visual over- or underrepresentation and propaganda misuse through red-teaming techniques. It is also internally testing a tool intended to detect images generated with DALL-E 3. These measures point to ongoing efforts to manage how the system can be used and how its outputs are recognized.