OpenAI is preparing to introduce the next generation of image creation models alongside the anticipated release of GPT-5.2. Two new models named Chestnut and Hazelnut have appeared on the test platforms LM Arena and Design Arena, which will likely be named Image-2 and Image-2-mini upon launch. They will succeed the popular Image-1 and focus on enhancing the quality and versatility of images.
Early users note significant improvements in detail and color accuracy in the new models. Issues with a yellow tint, which previously occurred in Image-1, have been resolved. This brings OpenAI’s results closer to the standard set by the Google Nano Banana 2 model.
Comparisons show that Nano Banana 2 still has an advantage in certain cases, but the new OpenAI models have narrowed this gap. The visual style of Chestnut and Hazelnut resembles early generations of DALL-E, but with noticeably better rendering quality. This allows for the creation of images for infographics, design, and marketing with more natural colors and structure.
The new models are planned to be integrated primarily into ChatGPT, and later into other OpenAI services where image creation is present. Preparations for the launch are already noticeable in the ChatGPT code, and the industry expects the announcement to coincide with GPT-5.2. These updates are expected to help creative professionals, developers, and companies obtain higher-quality visual materials for their projects.
