In episode 196 of The Presentation Podcast, Rolly Seth, Principal Product Manager for Microsoft Designer, made an interesting comment that written language is challenging for AI image generators. This was a thought-provoking insight, and yesterday it had me thinking about my experience experimenting with Copilot’s image creation. While the tool can proofread my emails and documents, there is no proofreading text in an image it creates.
Creative tools have undoubtedly changed, but the AI tools still seem fragile and unreliable at this stage. Even the ability to iterate an image and make adjustments to it is non-existent. While I’m not discounting this amazing tool, it’s not yet ready to be my go-to tool for design work. As example, in this almost great image, the word that is literally spelled out in the prompt was misspelled.
For reference, the prompt that Copilot (ChatGPT-4) used to generate the image was, “Create an image of a person pushing a button, labelled “present””.
Troy @ TLC