Last week, Microsoft published an article about AutoGen Studio, showcasing its ability to create a complete children’s book, including the story and illustrations, using a simple prompt. This video replicates those steps to create a small picture book using AutoGen agents, complete with images and descriptions.
The process starts by installing AutoGen Studio with Pip and then launching it using the command ‘autogen studio UI,’ which opens the application on Local Host Port 8081. After creating a GPT-40 model and testing it with an API key, the necessary skills for generating images and PDFs are added to the agent. The user then creates an assistant agent named ‘book agent,’ assigns the GPT-40 model to it, and adds the skills for generating and saving images and PDFs.
A new workflow named ‘book workflow’ is created, with the user proxy and book agent assigned to it. In the playground, the user instructs the agent to create an example image of an AutoGen AI agent with a short description. The console shows the internal communication between agents, and the result is an image with a description.
Next, the user requests the creation of a five-page PDF. After some waiting time, the agent produces a PDF, but the initial result lacks images. The user refines the request to include images in the PDF. The final result shows a PDF with images and text, but the images are placeholders, indicating areas for key features and use cases without detailed descriptions.
While the initial results are impressive, more time and refinement are needed to create a perfect picture book or brochure using AutoGen Studio. The video invites viewers to comment if they want a follow-up video with detailed steps to achieve this.