DALL-E 3 Capabilities Are Now Available in Bing Chat and Bing Image Creator

Microsoft has announced that DALL-E 3, the latest version of OpenAI's state-of-the-art text-to-image model, is now available in Bing Chat and Bing Image Creator for free.

DALL-E 3 represents a major leap forward in text-to-image AI, capable of producing strikingly realistic and diverse images from natural language prompts. Microsoft stated that over 1 billion images have already been created using Bing Image Creator since its launch. The integration with DALL-E 3 is expected to further boost users' creativity.

DALL-E 3 improves upon previous versions in several ways. The new model delivers higher overall image quality with greater detail, especially for human faces, hands, and text. Prompts are also followed with increased precision and reliability. Microsoft recommended providing very detailed prompts to get the most refined results from the AI system.

In addition to realism, Microsoft touted DALL-E 3's ability to generate logically coherent images that match prompts creatively. The company said DALL-E 3 goes beyond visual appeal to produce images with unique styles tailored to users' imaginative needs, whether crafting illustrated stories, social media posts, or other projects.

Microsoft also outlined measures implemented in Bing to ensure responsible and ethical AI practices. All AI-generated images include invisible digital watermarks certifying their origins per the Content Credentials for Provenance Standard. Additionally, a content moderation system removes harmful or explicit content that violates Microsoft's policies.

Here are some examples of some of the images we were able to generate:

Midjourney vs DALLE-3

We will be doing an in-depth comparison of DALLE-3 and Midjourney 5, but for now, here's a quick example of how they perform with the same prompt. This of course is not a fair comparison, since both models have their own jargon and interpret and process prompts differently. So rather than simply using the same prompt across both models, it will be better to learn the nuances, strengths and weaknesses of each model and adapt your prompts accordingly.