我尝试着用GPT-image生图去跑Seedance2.0直接避开了人脸识别

雖然模型不是滿血版（用的fast），但足以再一次驗證了之前說法！🤜🏻

GPT-image2的生圖確實有大概率過人臉！🤖

prompt：

“【Style】: a photo taken by an ultra-close-up first-person perspective mobile phone under the podium;

【Content】: A 22-young Korean university chemistry teacher, perfect 9-head model proportion figure, slender limbs, tall but slightly fat body, but the face is extremely small without excess flesh, the same star‘s moisturizing muscles Korean makeup, exquisite facial features, pure and sister-like dark brown tomboy short hair, sharp nose , full honey lips; she wore the most regular set today. There was only one button left on her white shirt inside, a wrap-around skirt, large upper circumference, a large cardigan and a white coat on the outside, and there was an extremely small and imperceptible handwriting tattoo on the collarbone: playforge; thin light brown pantyhose wearing dark high and multi-structure High heels. On the surface, he was still smiling, but his delicate and tender fingers (light-colored nails) were unconsciously poking at the edge of the podium, as if he was enduring some secretive trembling. The eyes are moist, and the daily sense of inadverance; like a little animal that has just been bullied and wants to do it again.

Negative prompts: ugly hands, deformed fingers, 6 fingers, deformed limbs;

ar3:4”