Several samples on X suggest that GPT-4o Imagegen can be used for infographics (1, 2, 3, 4). In my experiments, instructions to Imagegen need to be on-point: simply supplying a Readme.md with all kinds of different notes on a code project and asking it to visualize the build process does not work - the model will even start to hallucinate random things. What works instead is to ask GPT-4.5 to build the prompt for Imagegen, and keep it simple: elaborate layouting instructions like this also overwhelm the model:
[…] Near the bottom or top, add a small list of runtime dependencies (libusb, libglib2.0, etc.) to illustrate what must be present on the target system. […]
Payload instructions that works for the base diagram:
Thin Client Build Flow:
1. Box: “Prerequisites (CMake, Clang, Libraries)”
2. Box: “Build using CMake/Clang”
Intermediate result:
This can be iterated on, like adding a headline. Also, a style reference in the form of an existing diagram can be passed. I like Leonie Monigatti’s distinctive style, so gave one of her diagrams as the style reference:
Redo the first chart in the style of the second. Do not change the content, just the style.
Further iterations:
- “Turn it into an infographic, Studio Ghibli stlye. Include an instructor character that’s pointing to the flowchart.”
- “Make the instructor look like me” (include photo)
- “Censor [some items]. For censoring, retain the original style and make it appear as if deleted with a rubber. Make it beautiful.”
Final result:
Food for thought: ethical implications and ethical conduct when using the works of others as a style reference.