-PRADA × YAKUZA 企划圣经 (v1.x)

-PRADA × YAKUZA campaign bible (v1.x) -10 cinematic film-still prompts per input -B&W + heavy grain + VHS on every output -Prada styling + Yakuza world/posture model choice: pick something with high prompt adherence, high cfg, i avoid "creative" models for this task. You want something that actually respects the style you're asking for. I used gpt-image-2 for this, worked perfectly. workflow: @ComfyUI. first goal → lock a seed you like. test 3–4 random seeds. compare with refs vs without refs. you'll see the difference immediately.

-PRADA × YAKUZA 企划圣经 (v1.x) -每个输入产出 10 条电影级剧照提示词 -每张输出均为黑白 + 强颗粒 + VHS 质感 -Prada 造型 + 黑道世界/姿态 模型选择：挑选提示词遵循度高、cfg 值高的模型，我在这类任务上避免使用"创作型"模型。你需要的是真正尊重你所要求风格的模型。我用 gpt-image-2 跑这个，效果完美。 工作流：@ComfyUI。第一步 → 锁定一个你喜欢的 seed。 测试 3–4 个随机 seed，对比有参考图和无参考图的结果，差异一目了然。