PROMPT RADAR / VIDEO视频叙事
holy.. Claude does motion design now 🤯 you describe the video in chat a
holy.. Claude does motion design now 🤯 you describe the video in chat and Higgsfield’s MCP handles the rest - model choice, the shots, the render, the finished file. the mechanism: one OAuth connection, 30+ video models
@dr_cintas 871 热度 1.5 万 浏览
编辑先看
它为什么值得进入灵感库
它最有价值的部分，是把一个模糊想法推进到了可以被模型执行的程度。 把一段描述拆成起始状态、动作变化、镜头响应和结束画面，连续性比形容词更重要。
01 / 原始方法
动作与镜头 Prompt
holy.. Claude does motion design now 🤯 you describe the video in chat and Higgsfield’s MCP handles the rest - model choice, the shots, the render, the finished file. the mechanism: one OAuth connection, 30+ video models inside Claude - Veo 3.1, Sora 2, Kling 3.0. you describe the shot, Claude picks the model and returns the finished clip. up to 4K, 15s. no key, no editor. how to set up (2 min): 1. → Settings → Connectors → Add custom connector 2. paste 3. sign in with your Higgsfield account - OAuth, no key then just ask: “make me a 9:16 clip of…” that’s the full setup. your buddy’s still stitching renders across four browser tabs. you type one sentence and the shot comes back done.
02 / 复刻路径
不要一次改完所有变量
用 Veo 复刻时，建议先单独生成稳定首帧，再把动作拆成“开始、变化、收束”三段；每段只保留一个主要运动。
- 主体：holy.. Claude does motion design now 🤯 you describe the video in chat and Higgsfield’s MCP handles the rest - model choice, the shots, the render, the finished file. the mechanism: one OAuth connection, 30+ video models
- 视觉方向：实时 AI 视频案例；查看原帖媒体与作者说明
- 镜头与构图：按原帖镜头流程执行；查看原帖媒体与工作流
- 交付约束：auto，保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
负向约束
保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
模型建议
Veo / Kling / Seedance。先用原始 Prompt 建立基准结果，再根据模型对自然语言、镜头运动或文字排版的能力做局部改写。
来源与编辑原则