ChatGPT Images 2.0 is better at rendering non-Latin text

Source: Engadget

Overview

OpenAI’s new ChatGPT Images 2.0 model is now available.
A little more than a year after OpenAI gave ChatGPT users the option to create images and designs directly from its chatbot, it is releasing Images 2.0. OpenAI describes the system as a “step change” for image‑generation models, especially in its ability to follow detailed instructions, render dense text, and place and relate objects within a scene. For the first time, the model includes reasoning capabilities, allowing it to search the web and verify its outputs. According to the company, these capabilities should make the tool more reliable when accuracy, consistency, and visual cohesion are essential.

Improved Non‑Latin Text Rendering

OpenAI reports significant gains in handling non‑Latin scripts such as Japanese, Korean, Chinese, Hindi, and Bengali. The model is also better at faithfully recreating the specific characteristics of different visual languages, making it useful for tasks like game prototyping and storyboarding.

Additional technical improvements include:

• Flexible aspect ratios (as wide as 3:1 and as tall as 1:3)
• Resolutions up to 2 K
• Ability to generate up to eight outputs in a single request

Demo Tests

I previewed Images 2.0 ahead of its public release with three prompts:

1. Pixel‑art tortoiseshell cat in the style of Pokémon’s third generation – a challenging test for AI models, which often struggle with the distinctive Game Boy Advance aesthetic.
   

2. Transparent PNG conversion – the model generated a proper transparent image, a task many image generators find difficult.
   (The transparent version is not shown here, but the output was successful.)

3. Four‑page manga about the cat enjoying a sunny day by an idyllic city stream – the resulting manga deviated slightly from the original cat design but demonstrated the model’s ability to produce multi‑panel storytelling.
   

Availability

Images 2.0 is available starting today for all ChatGPT users, including those on the Free and Go tiers. Plus and Pro subscribers receive access to more advanced outputs. OpenAI is also offering the model through its API service and the Codex coding app, which was recently updated to include built‑in image generation.

The release comes shortly after Anthropic entered the visual‑design market with its own design assistant.

Relevant links:

• Generate images directly from ChatGPT (Engadget)
• Google’s Nano Banana 2 (Engadget)
• OpenAI’s Go tier details (Engadget)
• Codex app update (Engadget)
• Anthropic design assistant (Engadget)