My GPT Image 1.5 vs. Nano Banana Pro Test: The Forbidden City as a 3D Scroll

Just as I expected, Nano Banana Pro does a much better job respecting historical accuracy and architectural logic, no surprise given it’s powered by Gemini 3 Pro and taps into Google Search’s knowledge base. 

GPT Image 1.5, on the other hand, completely violates the most fundamental principle of Chinese imperial palace design, especially for buildings of the Hall of Supreme Harmony’s caliber: all principal halls must align along a single north-south axis, with strict bilateral symmetry. On top of that, every single Chinese character it rendered was wrong. Guess @OpenAI called this update “1.5” instead of “2” because they know full well the model gets crushed by 🍌 when it comes to world knowledge and reasoning.

Still, for all its shortcomings in common sense and reasoning, I really do love the creativity GPT Image 1.5 brings. The flowing scroll, the misty atmosphere, the bolder palette, it all feels like a scene straight out of the ink-wash animations I’ve always imagined.

This comparison once again confirms what makes ChatGPT truly irreplaceable: its intuitive grasp of creative elements, its delicate touch, its refusal to be generic. But in the mad dash to chase benchmarks, OpenAI keeps throwing these qualities away with each new release, along with the users who loved ChatGPT for exactly these reasons.

Not that I expect them to care. After all, they just inked a billion-dollar deal with Disney. As long as new features keep going viral, they’ll give zero thought to what made ChatGPT great in the first place.​​​​​​​​​​​​​​​​

#StopAIPaternalism @sama @gdb @nickaturley @fidjissimo