ChatGPT Images 2.0: A Revolution in Text-to-Image Generation

⚡

Key Takeaways

1ChatGPT Images 2.0 surpasses its predecessors by generating menus without text errors.

2OpenAI improves the accuracy of non-Latin texts, covering Japanese and Korean.

3Images 2.0 offers a resolution of up to 2K, allowing for detailed and complex creations.

💡Why it matters — This advancement strengthens the use of AI in marketing and visual content creation, expanding its professional applications.

The Impressive Evolution of AI Image Generation

A few years ago, images generated by artificial intelligence were easily identifiable by their glaring errors, particularly in text reproduction. For instance, trying to create a menu for a Mexican restaurant with these tools often resulted in fanciful creations such as "enchuita" or "churiros," non-existent dishes that immediately betrayed their artificial origin.

Today, with the ChatGPT Images 2.0 model, that era seems to be over. When asked to design a menu of Mexican dishes, the result is precise enough to be used directly in a restaurant, without customers suspecting anything. While the price of a ceviche at $13.50 might still raise some questions about the quality of the fish offered, the spelling and presentation are impeccable.

Comparison with Previous Generations

To better understand this advancement, it is helpful to recall the limitations of earlier models like DALL-E 3. Two years ago, these image generators could not yet produce correct text, as they relied on diffusion models. These models reconstruct images from noise, making the faithful reproduction of text particularly difficult.

Asmelash Teka Hadgu, founder and CEO of Lesan AI, explained in 2024 that these models focused on patterns covering the majority of pixels, thus neglecting small portions of text. This approach limited their ability to generate accurate text.

New Approaches and Technologies

To overcome these challenges, researchers have explored other methods, including autoregressive models. These models predict the appearance of images sequentially, somewhat like language models such as LLMs (Large Language Models).

However, OpenAI has not specified what type of model exactly powers ChatGPT Images 2.0. At a recent press conference, the company declined to answer this question, leaving the technical details of their new technology shrouded in mystery.

Advanced Capabilities and Applications

Nevertheless, OpenAI revealed that the Images 2.0 model incorporates reflective capabilities. This means it can perform web searches, generate multiple images from a single request, and verify its own creations. These features allow for the production of marketing materials of various sizes or even multi-panel comics.

The model is also capable of better understanding and reproducing text in non-Latin languages, such as Japanese, Korean, Hindi, and Bengali. However, its knowledge base stops in December 2025, which could limit its accuracy for more recent events.

Unmatched Level of Detail

According to OpenAI, Images 2.0 offers unprecedented specificity and fidelity in image creation. It can not only conceptualize sophisticated images but also execute them with great precision. The model follows the given instructions, preserves the requested details, and renders complex elements such as fine text, iconography, or dense compositions, all up to a resolution of 2K.

While generating complex images takes a bit longer than simple text requests, creating a multi-panel comic only takes a few minutes.

Accessibility and Pricing

All users of ChatGPT and Codex will have access to Images 2.0 starting Tuesday. Paying users will benefit from more advanced results. OpenAI also offers an API, gpt-image-2, with pricing varying based on the quality and resolution of the generated images.