GPT-6 Astra 与 Gemini 4 Pro 同题生图对比
同一提示词下 Astra 偏重时代氛围、Gemini 4 Pro 偏重物体细节，作者认为差距已不在画质而在对提示词的理解。
🚨 GPT-6 ASTRA VS GEMINI 4 PRO WERE ASKED TO SHOW THE SAME MOMENT IN HISTORY, AND THEY DISAGREED ON WHAT THAT MOMENT LOOKED LIKE First scene: a caveman inventing the wheel. Astra keeps it primal, a solid boulder, raw muscle, dust and dry grass, nothing refined about it yet. Gemini 4 Pro jumps ahead in the story on its own, the stone already has a center hole cut through it, closer to an actual wheel than a rock being pushed. Same prompt, different interpretation of "the moment." Second scene: a vintage motorcycle. Astra's version sits in front of an industrial skyline, smokestacks pumping in the background, like the bike belongs to the birth of the machine age. Gemini 4 Pro strips the environment down to open plains and empty sky, all attention on the mechanical detail of the engine and frame instead of the setting around it. Neither model is technically wrong here. They're just answering a different question hidden inside the same prompt, one treats it as "show me the era," the other treats it as "show me the object." That's the actual interesting part of these side by sides lately. The gap isn't image quality anymore, both of these are clean. It's that two models can read identical words and walk away with two different stories. Disclaimer: this footage may be staged, edited, or AI-generated, its origin is unconfirmed.
这条是引用着另一条写的：x.com/aiboticssq/status/2100916466694414727