ChatGPT’s new Images 2.0 model is surprisingly good at generating text

Text legibility in AI-generated images has historically been the tell that exposes synthetic content at a glance, so closing that gap has real consequences for detection, trust, and misuse surface area — none of which TechCrunch's framing addresses.

This lands in the middle of an active multimodal arms race. On April 16th, Google extended Gemini's image generation to pull from personal photo libraries, deepening its contextual generation story. That same week, OpenAI's Codex upgrade added image generation as one of its new agentic capabilities, per The Verge's coverage. Images 2.0's text rendering improvement is another piece of the same competitive push: both labs are stacking multimodal capabilities quickly, and the differentiation is shifting from 'can it generate images' to 'how precisely can it handle structured visual content like text, diagrams, and personal context.' OpenAI is clearly trying to make its image stack harder to dismiss as a secondary feature.

Watch whether Adobe, Canva, or any major design platform announces a direct integration with Images 2.0 within the next 90 days. A commercial design partnership would confirm that the text rendering gains are production-grade rather than benchmark-grade.