On December 10th, the source @marmaduke091 today (10 December) posted tweets on platform X, reporting that seven months after the gpt-image-1 release, it was not possible to get the message out of the mediaOpenAI It's being tested for "Hazelnut" and "Chestnut" AI IMAGE MODEL.
The explosion suggests that OpenAI is increasing the pace of product overlay, and that these models have been frequented in model arenas such as Design Arena and LM Arena in the last 24 hours。
If the news is true, OpenAI is likely to launch these two new models in the near future, or even as a companion product to the forthcoming GPS-5.2 model in the rumour。
THERE IS NOW WIDESPREAD SPECULATION WITHIN THE INDUSTRY THAT THE GPT-5.2 DISTRIBUTION WINDOW IS NEAR, AND THE UPDATING OF IMAGE GENERATION CAPACITY IS OFTEN AN IMPORTANT PART OF THE UPGRADING OF LARGE VERSIONS OF THE MODEL. HOWEVER, NO FINAL OFFICIAL CONFIRMATION OF THIS HAS YET BEEN MADE, AND A SPECIFIC RELEASE TIMETABLE IS YET TO AWAIT FURTHER INFORMATION。
Early test feedback indicates that the greatest technological breakthrough of the new model is the ability to render “image code”. The new model is more reliable in presenting a clear and accurate code text in the image than in the case of the previous model, which often produces a obscurity in the generation of images containing text or code。
In addition, the level of “world knowledge” that it displays has increased significantly and has been able to resist the understanding of the generic scene with the Google-based Nano Banana Pro Chamber. 1AI attaches the following pictures:
However, the media noted that, despite some breakthroughs in text rendering, the new models still needed to be sharpened in the quality of human creation。
SEVERAL X PLATFORM USERS, AFTER LOOKING AT THE TEST SAMPLE, POINTED OUT THAT, ALTHOUGH THE RESULTING CELEBRITY SELF-IMAGE WAS VERY REAL ON THE DRAWINGS, IT WAS NOT THE FIRST TIME THAT THEY WERE ABLE TO SEE THE RESULTSBut the facial skin texture shows a clear "plasticity"Lack of a real skin nuance. In contrast, the competitor's Nano Banana Pro still has a lead in human detail and nature。
In terms of technical architecture, industry hearsay that these two models are being tested not on the basis of the latest gpt-image-1, but on the basis of the early GPT-4o model development, which may mean that OpenAI is trying to optimize image generation through different technological pathways. Of course, before the official release of the technical white paper, the details of the bottom structure of the model were still in the speculative stage。