OpenAI has released ChatGPT Images 2.0, a new image generation model that significantly improves text rendering and visual accuracy, addressing long-standing limitations that previously made AI-generated images easy to identify.
Improved Accuracy In Text And Visual Details
Earlier image models often struggled with spelling and text placement, producing distorted words in generated images. Two years ago, models like DALL-E 3 frequently generated incorrect text when asked to create items like restaurant menus.
ChatGPT Images 2.0 now produces text that appears accurate enough for real-world use. In testing, a generated Mexican restaurant menu contained correctly spelled items and coherent formatting, reflecting a shift in how image models handle fine details.
Shift In Underlying Model Capabilities
Previous image systems relied heavily on diffusion models, which reconstruct images from noise and tend to prioritize large visual patterns over smaller elements like text.
As explained by Asmelash Teka Hadgu, diffusion approaches often underrepresent written elements because text occupies only a small portion of image pixels.
Newer approaches, including autoregressive methods, allow models to predict image components more sequentially, similar to how large language models generate text. OpenAI did not disclose the exact architecture behind Images 2.0.
Introduction Of “Thinking Capabilities”
OpenAI said the new model includes “thinking capabilities,” enabling it to perform tasks such as generating multiple images from a single prompt, checking its own outputs, and incorporating external context through web searches.
These capabilities support more complex outputs, including multi-panel comic strips, marketing assets in different formats, and images that maintain consistent structure across variations.
Expanded Language And Design Support
The model shows improved handling of non-Latin scripts, including Japanese, Korean, Hindi, and Bengali. OpenAI said Images 2.0 can better render detailed elements such as small text, iconography, user interface components, and dense visual compositions.
The system supports outputs at up to 2K resolution and is designed to follow detailed instructions while preserving stylistic constraints.
Availability And Access
ChatGPT Images 2.0 is rolling out to users of ChatGPT and Codex starting Tuesday. Paid users will have access to higher output limits and more advanced generation features.
OpenAI is also releasing the gpt-image-2 API, with pricing based on output quality and resolution. The model’s knowledge cutoff is December 2025, which may affect prompts related to recent events.
Featured image credits: RS Gonzales
For more stories like it, click the +Follow button at the top of this page to follow us.