OpenAI Images 2.0: Image generation with Thinking Apfelpatient

OpenAI has released a completely redesigned image generation model: ChatGPT Images 2.0 combines "thinking" capabilities with significantly improved text rendering and real-time web search. The model is now available to all ChatGPT, Codex, and API users.

OpenAI continues to expand its product portfolio. Following the recent release of GPT-5.4 and GPT-5.4 mini and nano, the next generation of image generation is now available: ChatGPT Images 2.0. This model is designed to handle significantly more complex visual tasks than its predecessor and features a new capability for AI image generation – the ability to incrementally re-evaluate results.

Image generation AI has been a competitive field for years – DALL-E, Midjourney, Stable Diffusion, and numerous competitors are vying for technological dominance. With Images 2.0, OpenAI is now attempting a fundamental leap: The model can, if necessary, check and correct its generation and even access the internet to retrieve current information. This is significantly more than classic text-to-image generation.

ChatGPT Images 2.0: What's new?

The most important new features of ChatGPT Images 2.0:

Thinking capabilities: The model can work step-by-step through complex instructions and correct itself. OpenAI expects this to lead to more precise results with difficult compositions.
Improved text rendering: Text in generated images should become significantly more legible and precise. This has been one of the biggest weaknesses of all image generation models to date.
Multilingual support: Non-Latin characters such as Japanese, Korean, Chinese, Hindi and Bengali are now displayed better.
Real-time web search: The model can search the internet for current information to generate images based on real data.
Multiple variations simultaneously: A single prompt can generate up to eight images at once – useful for comparing different interpretations.
Multiple aspect ratios: Images can be created in different formats.
2K resolution: The maximum resolution has been increased to 2K.

"Thinking" – what does that mean in concrete terms?

The most important step forward is the "thinking" capability. Previous image generation models received a prompt and immediately output an image. Images 2.0, on the other hand, can go through several processing steps, evaluate the output itself, and make improvements if necessary.

In practice, this could mean that if a user requests an image with specific objects in a particular arrangement, the model checks whether all objects are indeed positioned correctly – and corrects itself if necessary. This solves one of the most persistent problems in AI image generation: the inaccurate execution of detailed instructions.

Improved visual appeal

An interesting phrase in OpenAI's announcement states that Images 2.0 has an "improved sense of composition and visual taste." The result should be images that appear less "AI-generated."

This is an often overlooked problem. AI-generated images often have a characteristic look that viewers immediately recognize as artificial – unnatural lighting, exaggerated details, certain compositional patterns. OpenAI is now trying to reduce this „AI look“ by having the model make more aesthetically sound decisions.

Whether this works in practice will be shown by initial tests. Realistically, AI-generated images will still be recognizable to trained eyes – but for average users and for commercial applications, the results could be significantly more convincing.

What this means for iPhone users

This update is relevant for Apple users because ChatGPT is deeply integrated into iOS. Users can interact with ChatGPT directly via Siri or use the new image capabilities in the ChatGPT app on their iPhone. Those who already subscribe to ChatGPT Plus or Pro will receive Images 2.0 at no extra cost.

The contrast to Apple's own AI image generation is striking. Apple Intelligence also offers image features with "Image Playground," but it appears significantly more limited compared to Images 2.0. Apple's future use of Gemini technology for Siri suggests that the company is also considering external solutions for image and text-to-image tasks.

OpenAI: The broader context

Images 2.0 joins an impressive output of new products from OpenAI in recent months. Alongside GPT updates and the new cybersecurity model GPT-5.4-Cyber, the company is also working on new areas such as the discontinued video model Sora. The speed at which OpenAI releases new products reflects the current pace of AI development – and demonstrates how difficult it is for Apple to keep up in this field.

The good news for regular users is that Images 2.0 gives ChatGPT subscribers a significantly more powerful image tool at no extra cost. For professional users—designers, marketers, and content creators—the improved text rendering and multilingual support should be particularly valuable. Anyone who has previously avoided AI-generated images due to unsatisfactory quality or imprecise implementation should definitely take a second look at Images 2.0. (Image: Shutterstock / Camilo Concha)