Microsoft unveiled the Image Creator, a new feature integrated into their AI-empowered Bing search engine, in March. The Image Creator can transform written descriptions into corresponding images, allowing users to explore an infinite array of image possibilities directly within Bing. Users can utilize this feature to create specific images they require rather than trawling the internet. This tool proves helpful in various situations, including conceptualizing a dream, designing a themed birthday invitation, or illustrating a new idea at work. The created images are personal, precise, and unique, enabling users to communicate their ideas. Bing Chat also enables users to request images during conversations.
A Closer Look at the Image Creator
Microsoft has offered insight into the technology and thought process that led to creating the Image Creator. Additionally, they have provided valuable tips to optimize the feature within the product.
The Image Creator leverages OpenAI’s advanced DALL·E model, a potent language model capable of generating coherent and varied text across diverse topics. Unlike typical language models that output text from the input text, DALL·E uses input text to output images. This is achieved through a method known as self-attention, which enables the model to learn the correlations between text and images.
This technology allows the Image Creator to generate realistic images that blend concepts, attributes, and styles. Users can ask Bing Image Creator to produce images based on descriptions such as a “cat wearing a hat and sunglasses” or a “surreal painting of a pineapple”. The more detailed the user’s query, the more specific the resultant image. Users are advised to specify a subject, details, and the scene to optimize results. Instead of only using the Image Creator interface, users can also generate images iteratively within a Bing chat session.
The Rationale Behind Microsoft’s Phased Release
Microsoft initially introduced the Image Creator tool in select markets in October 2022, followed by a wider release in March. The company has also clarified the reasons for the staggered rollout.
Microsoft was committed to understanding the responsible use of AI and wanted to conduct comprehensive testing. While the Image Creator uses OpenAI’s advanced DALL·E model to generate AI images, Microsoft also prohibits the tool from creating potentially harmful content to individuals or society.
To ensure this, the company has established measures to block specific prompts and images, utilizing the months leading to the broader rollout to refine this strategy. Users who feel their images are unjustly blocked can appeal, thereby helping Microsoft refine its classifiers.
Microsoft expects to receive an improved model from OpenAI that will offer better image quality, a detailed understanding of visual concepts, and thus more pertinent results. Users can test the Image Creator for free in English, wherever the new Bing is available.