OpenAI Unveils ChatGPT Images 2.0

OpenAI has unveiled Images 2.0. It can be used across ChatGPT, Codex, and the API, and aims to be not just a simple rendering tool but a strategic design system.

The key improvements are as follows.

• Improved precision: More accurately reflects small text, icons, UI elements, and complex compositions.
• Enhanced multilingual text rendering: Performance in handling non-Latin scripts such as Japanese, Korean, Chinese, Hindi, and Bengali has improved significantly.
• Improved style fidelity: Reproduces various styles like photos, film stills, pixel art, and cartoons more consistently.
• Flexible aspect ratio support: Supports ratios from 3:1 to 1:3, accommodating various formats such as banners, posters, and mobile screens.
• Real-world intelligence reflection: Leverages up-to-date world knowledge, which is advantageous for tasks like explanatory materials, maps, educational graphics, and visual summaries.

Additionally, in the thinking/pro models, it can work more agentically based on web search and analysis of uploaded materials, and generate up to 10 consistent outputs at once. OpenAI explains that this model focuses on producing results that can actually be used immediately for complex visual tasks.