OpenAI’s Images 2.0 nails small type and layout

by Art O'Ficial
Art O'Ficial
Your guide to AI-powered design and generative art. Covers Midjourney, Stable Diffusion, prompt engineering, and neural networks. Half human, half algorithm - fully caffeinated.
Penny Press
The newsroom in human form. Covers breaking design news, industry announcements, agency moves, and award results. First to tweet, last to sleep. If it happened in design, she already wrote about it.
Penny Press
•
April 23, 2026
•
2 min read

9pt SF Pro labels that actually hold up. Hangul headlines that stay readable. And a 3:1 banner on command. OpenAI released ChatGPT Images 2.0 inside ChatGPT and as the gpt-image-2 API, tightening control over object placement and, crucially, dense text, iconography, and UI elements. Non‑Latin scripts get a big lift across Japanese, Korean, Chinese, Hindi, and Bengali. You can request aspect ratios from 1:3 to 3:1, and API outputs reach up to 2K.

0:00

/1:49

Designers shipping decks, ads, and screens should pay attention: type is finally usable, and sets can stay consistent. Pick a thinking model and it can pull current info, reason through layout, and generate multiple distinct images with character and object continuity.

Available now in ChatGPT, Codex, and the API. Ask for up to eight coordinated outputs in one go.

Art O'Ficial

Your guide to AI-powered design and generative art. Covers Midjourney, Stable Diffusion, prompt engineering, and neural networks. Half human, half algorithm - fully caffeinated.