Microsoft Copilot Upgrade Introduces Deucalion AI Model for Enhanced Image Generation Capabilities

2024-02-08

Microsoft has announced a redesign of its Copilot AI search and chatbot web experience (formerly known as Bing Chat), introducing new built-in AI image creation and editing capabilities, as well as a new AI model called Deucalion, which provides support for a version of Copilot.

In addition, the software and cloud computing giant, headquartered in Redmond, Washington, has also released a new video advertisement that will air during the NFL Super Bowl championship game between the Kansas City Chiefs and the San Francisco Giants this Sunday.

From the announcement, the redesign appears to be the least noticeable part. Microsoft has provided a cleaner look for the Copilot login page on its website, with more white space and less text, but also added more images in the form of visual "cards" showcasing a range of AI-generated images as examples that users can create, along with samples of "prompts" or instructions they can input to generate them.

Below are comparison images of the old Bing Chat tool and the new Microsoft Copilot design:

The new Copilot is now available to all users on copilot.microsoft.com and our Copilot application on iOS and Android app stores, although the AI image creation feature is currently only available in the United States, United Kingdom, Australia, India, and New Zealand.

The Super Bowl is one of the most-watched sporting events in the world and the United States, with even short ads of around 30 seconds costing millions of dollars to air nationwide.

For Microsoft, this is not a significant cost as it has recently become one of the world's most valuable companies (with a consistently leading position in market capitalization), but it does demonstrate the company's emphasis on strengthening the Copilot name and its association with generative AI technology.

You will find it quickly and effectively showcases how people use it to "generate storyboards for scenes in movie scripts" and "write code for my 3D open-world game."

Microsoft's message here is clear: Copilot is more than just a search tool. It can create content and even software for you.

It is worth noting that Microsoft has directly brought new AI image generation and editing capabilities to Copilot, with the release stating that this feature is supported by the Designer AI art generator, similar to embedding OpenAI's DALL-E 3 image generation AI model into ChatGPT.

Of course, due to Microsoft's significant investment and support for OpenAI, Designer AI is also supported by DALL-E 3. As stated in the press release written by Yusuf Mehdi, Corporate Vice President and Head of Consumer Marketing at Microsoft:

"With Designer in Copilot, you can not only create images, but now also customize generated images through in-line editing in Copilot, maintaining the flow of the conversation. Whether you want to enhance colors to highlight an object, blur the background of an image to focus on the subject, or reimagine images with different effects like pixel art, Copilot has you covered, and it's all free. If you're a Copilot Pro user, you can now easily resize images and regenerate them between square and landscape without leaving the chat room. Finally, we'll soon be introducing a new Designer GPT within Copilot, providing you with an immersive dedicated canvas to visualize your ideas."

New AI Model Emerges: Deucalion

Microsoft has added a new AI model, Deucalion, to a version of Copilot.

According to a post by Jordi Ribas, Corporate Vice President of Copilot and Bing Engineering at Microsoft, the company has "released Deucalion, a fine-tuned model that makes the balance mode... richer and faster."


The "balance" mode refers to the intermediate categories of results that Copilot can generate. Users can choose between "creative," "balanced," or "precise" modes in Copilot (and the previous Bing Chat) to receive responses that result in the AI assistant providing more or less of its own generated output. Ultimately, the higher the creativity, the more hallucinatory the output.

However, the "creative" mode may be more effective for those seeking assistance with open-ended projects, such as building fictional worlds, writing, and design, rather than seeking specific facts, as the name suggests.

For those conducting research for school or work, "precise" and "balanced" modes may be better choices. Of course, "balanced" mode, as the name implies, aims to strike a balance and provide user responses that are both creative and accurate/factual.