ChatGPT Images 2.0 Has Been Released

ChatGPT Images 2.0 has been released. Image Arena is already recording a historical lead over competitors. The understanding of physics and space is excellent, but users often encounter fractal artifacts. It is rumored that this is a bug of the specific version and it may be fixed with an update after the first wave of tests. The main feature of the release is 'thinking capabilities'. Essentially, this is the first nanobanana-like model from OpenAI that really thinks before generating. It searches for information itself with a cutoff in December 2025, builds a composition, and produces up to 8 consistent images per prompt. The signature 'yellowing' has not gone anywhere, and people's skin turns out too perfect and glossy (we solve this in GenPeach AI 😁). In terms of photorealism, Banana still performs better. However, the text generation is simply outstanding — the model excellently depicts small details, so UI interfaces turn out great. Apparently, they loaded half of the internet as usual. Other perks include any aspect ratios and 2K generation via API (but not in 4K, after all). So we generate the interface in the new gpt-image-2, feed it into Claude Design for translation into HTML, and another million-dollar business is ready; buy my courses.