步骤 4：

Step 4: Generate a start frame using GPT-Image-2. My prompt and ref image: A screenshot from a live Premier League football (soccer) match TV broadcast on Sky Sports. The camera cuts to the crowd — our reference image person, sitting smiling. Smiling naturally, unaware they're on camera. The subject is sitting in a premium pitchside / front-row seat in the stands behind the touchline, stadium crowd visible around them. Hardlock: Do not alter their facial structure and maintain their likeness. Full Sky Sports broadcast overlay: scorebug in the top-left with team crests, match clock, score, and competition badge; Sky Sports network logo watermark in the corner; lower-third graphics strip; 16:9 aspect ratio. The image looks exactly like a real TV screenshot — broadcast color grading, slight compression artifacts, interlacing grain, the saturated green of a floodlit pitch bleeding warm light into the stands. It's Arsenal vs Tottenham in the second leg of the FA Cup semi-final at the Emirates Stadium. Score reads ARS 2 – 1 TOT, 67th minute. Arsenal lead 3-1 on aggregate. Evening kickoff, floodlit, packed stadium atmosphere.

步骤 4： 使用 GPT-Image-2 生成一个起始帧。 我的提示词和参考图像： 一张来自天空体育（Sky Sports）直播的英超足球比赛的电视转播截图。镜头切到观众席——我们的参考图像人物坐在那里微笑。自然地微笑着，没有意识到自己被镜头拍到。主体坐在靠近球场边/前排的优质看台座位上，位于边线后方的看台中，周围可见体育场内的观众。硬性锁定：不要改变他们的面部结构，保持他们的外貌特征。完整的天空体育转播叠加层：左上角的记分牌，包含球队队徽、比赛时钟、比分和赛事徽章；角落里的天空体育台标水印；下三分之一处的字幕条；16:9 宽高比。这张图像看起来就像一张真实的电视截图——广播级的调色、轻微的压缩伪影、隔行扫描颗粒感，泛光灯照射下球场的饱和绿色将温暖的光线漫射到看台上。这是阿森纳对阵托特纳姆的足总杯半决赛第二回合，比赛在酋长球场进行。比分显示 ARS 2 – 1 TOT，第 67 分钟。阿森纳在总比分上以 3-1 领先。傍晚开球，泛光灯照明，座无虚席的体育场氛围。