ChatGPT users are generating 1.5 billion images per week. OpenAI product lead Adele Li and researcher Kenji Hata join host Andrew Mayne to break down what changed with Images 2.0, covering specific capability jumps: text rendering, photorealism, multilingual support, character consistency, and flexible aspect ratios.
The episode is worth reading in full for the technical sections, particularly the training breakthroughs behind photorealism starting at 10:51, and the evals and prompting discussion at 14:06, where the team gets specific about how creative control actually works under the hood. The 360-image use case at 05:25 is an early signal of where spatial and productivity applications are heading.
The final section, Images plus Codex at 22:27, is the one to watch. It points at image generation moving from a standalone tool toward integration with code generation, which changes what the product actually is. Prompt tips close the episode at 28:08 for anyone who wants immediate practical use.
[WATCH ON YOUTUBE →]