"生成一张高度逼真的抓拍照片,看起来像是朋友用普通智能手机随手拍摄的。

image prompt: "generate a highly realistic candid photo that looks like it was taken by a friend using a normal smartphone. the subject is a woman in her 20s sitting at a small restaurant at night with her friends. the photo captures the exact moment she notices the camera and casually turns her head toward it. she has a natural expression with a slight look of surprise. her face is clearly visible. the restaurant has warm yellow lights, wooden furniture, food plates, drinks, and menu boards in the background. the image should look like a real frame taken from someone's instagram reel or tiktok video. include natural skin pores, tiny facial imperfections, realistic hair, slightly uneven exposure, subtle motion blur, and imperfect camera framing. make it feel spontaneous, ordinary, and completely unstaged. avoid the look of a professional photoshoot or a polished commercial. negative: artificial skin, plastic face, 3d, cgi, illustration, cartoon, anime, beauty retouching, excessive glow, perfect lighting, unnatural eyes, strange hands, forced smile, overly symmetrical face, cinematic lighting, heavy background blur, over-sharpening, studio portrait, unrealistic environment." and here's one more thing. stop adding words like "masterpiece," "perfect lighting," and "flawless beauty" to every prompt. those words can push your images toward that polished AI look you're trying to AVOID. you're not trying to make a perfect image. you're trying to make a BELIEVABLE image. that's a MASSIVE difference. left: OpenAI gpt images 2.5 right: google's nano banana 2.1

"生成一张高度逼真的抓拍照片,看起来像是朋友用普通智能手机随手拍摄的。 主体是一位二十多岁的女性,夜晚与朋友们坐在一家小餐厅里。 照片捕捉到她注意到镜头、并随意转头看向镜头的那个瞬间。 她表情自然,略带一丝惊讶,面部清晰可见。 餐厅背景有温暖的黄色灯光、木质家具、餐盘、饮品和菜单板。 图像应该看起来像是从某人的 instagram reel 或 tiktok 视频中截取的真实一帧。 包括自然的皮肤毛孔、细微的面部瑕疵、真实的发丝、略微不均匀的曝光、轻微的运动模糊,以及不完美的取景构图。 让画面感觉是自发的、平凡的、完全未经摆拍的。 避免专业摄影棚拍摄或精致商业广告的感觉。 负面提示:人工皮肤、塑料脸、3d、cgi、插画、卡通、动漫、美容修图、过度光晕、完美打光、不自然的眼睛、奇怪的手、强迫的笑容、过度对称的脸、电影感打光、强烈的背景模糊、过度锐化、影棚人像、不真实的环境。" 还有一件事。 停止在每个提示词里添加"masterpiece"、"perfect lighting"、"flawless beauty"这样的词。 这些词会把你的图像推向那种你正试图避免的精致 AI 风格。 你不是在试图制作一张完美的图像。 你是在试图制作一张**可信**的图像。 这是一个巨大的区别。 左侧:OpenAI gpt images 2.5 右侧:google's nano banana 2.1