OpenAI has released GPT-Image-2, integrated into ChatGPT as Images 2.0, which significantly improves the quality and accuracy of AI-generated images, especially in rendering text. Unlike earlier models like DALL-E 3, which often produced nonsensical text in images, Images 2.0 can generate realistic and usable visuals such as a Mexican restaurant menu without obvious errors, although minor details like pricing might raise questions.
Advances in Image Generation Technology
Traditional AI image generators relied on diffusion models that reconstruct images from noise, often struggling with small details like text. Asmelash Teka Hadgu, CEO of Lesan AI, explained that diffusion models focus on larger image patterns, making text a tiny and difficult part to replicate accurately. To overcome these limitations, researchers have explored autoregressive models, which predict image content more like large language models (LLMs), improving text generation within images.
New Features and Capabilities of Images 2.0
OpenAI’s Images 2.0 model includes “thinking capabilities” that enable it to:
- Search the web for information,
- Generate multiple images from a single prompt,
- Double-check its outputs for accuracy.
These features allow the creation of complex marketing assets in various sizes and multi-paneled comic strips. The model also supports better rendering of non-Latin scripts such as Japanese, Korean, Hindi, and Bengali. However, its knowledge base cuts off in December 2025, which may affect the accuracy of images related to recent events.
Quality and Performance
OpenAI highlights that Images 2.0 offers unprecedented specificity and fidelity, capable of following detailed instructions and preserving fine elements that typically challenge image models. These include:
- Small text,
- Iconography,
- UI elements,
- Dense compositions,
- Subtle stylistic constraints.
The model can generate images up to 2K resolution. While image generation is slower than text generation in ChatGPT, creating complex visuals like multi-paneled comics takes only a few minutes.
Availability and Access
Images 2.0 will be accessible to all ChatGPT and Codex users starting Tuesday, with paid users able to generate more advanced outputs. OpenAI will also release the gpt-image-2 API, with pricing based on output quality and resolution.
