As we already reported, it seems OpenAI is preparing to release new image generation models.
Honestly, despite the fact that the current model used in ChatGPT appeared quite recently and quickly became popular, today it already looks a bit outdated. Competitors are not sleeping and almost every week release something new — with new features, improved color rendering, and overall visual quality.
Therefore, the potential appearance of the next generation of models is definitely good news. If, of course, the information is confirmed. There is currently no exact data on whether these unknown models are exactly Image-2 and Image-2-mini, but everything points to this.
So far, they can only be tested on online arenas. And it’s not very convenient — you have to run comparisons with other models again and again, hoping that the new ones will appear among the candidates. However, I tried.
And here’s what came out:
First impressions are as follows:
The visual has become noticeably more diverse. The current model often adds a well-known yellowish “old” tint and recognizable style. In contrast, the new results look much more varied — as if the model has abandoned template-like behavior and allows for more variability.
The detailing has also changed. Literally everything is immersed in small elements, textures, and nuances — sometimes even excessively, but it’s better than the opposite.
There is often a blurred background or slight blurring of the object in motion. Perhaps this is how the model tries to add realism and “camera-like” quality to the frame.
Another interesting detail — some images resemble a refined DALL-E with more mature rendering and cleaner structure.
Of course, much can change before the official release, so we will monitor how the situation develops. But it is already clear that this is potentially a big step forward. And possibly the very upgrade that OpenAI has long owed its ecosystem.
