Insights on why OpenAI’s 4o Image Generation is a BIG deal

“Our servers are melting”, Sam Altman. Insights on why OpenAI’s 4o Image Generation is a BIG deal and a hack to get coax images out of OpenAI.

Two months ago I didn’t produce a banner image for my newsletter out of frustration working with diffusion tools like DALL-E. Getting what I wanted was difficult and random.

The recent launch of autoregressive image generators by Google and OpenAI provides unprecedented control with creativity, including correctly rendering text. This has major implications. To learn more, here are some helpful links:

Ethan Mollick provides a good overview (upside and downside) and examples with prompts. link

Jon Victor, in Applied AI, outlines some quick use case examples, including: staging empty apartments and creating building renderings, mocking up websites, designing branded merchandise, and—one I will use—building custom charts and graphics. link

Nathan Lambert’s piece dives into the technical aspects of this new image generation, what ‘autoregressive’ means, and the difference between launching technology (Gemini came out first, but access is too technical) and products (OpenAI’s approach, two weeks later). link

Soups Ranjan points out the fraud implications such as falsifying records and insurance claim trickery. link

My simple hack to get image our of OpenAI has been to check both chatgpt.com and the iPhone app. I’ve found images started on the website rendering more quickly in the app.