菜单
Use when generating a single talking-head or dialogue shot from a first frame, a minimal motion tag, a clean voice audio file, and dialogue text. This skill is for voice-locked shot rendering in the xiaolagumanju pipeline.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when converting character analysis results into Banana JSON prompts for text-to-image generation. This model benefits from explicit detail, reference-sheet layout, and strong negative constraints.
Use when converting shot-design results into Banana 3x3 storyboard grid JSON prompts for the current flashxiaolagu mainline. This skill is assembly-only and should write short natural-language cinematic beats, not rigid keyword chains.
Use when converting prop analysis results into Banana JSON prompts for the current Banana/Veo pipeline. This stage should preserve visible structure, layout blocks, multi-view detail, and generation-critical markings.
Use when converting scene analysis results into Banana JSON prompts for the current Banana/Veo pipeline. This stage should lead with one decisive narrative identity sentence, then give layered scene control rather than one giant brute-force room dump.
Use when one or more split storyboard panel images such as `storyboards/frames/.../panel_004.png` need Nano Banana Pro image editing instead of rerunning the whole Banana storyboard batch. This skill writes structured Chinese edit prompts for panel-level character, scene, prop, or fused repairs while explicitly preserving locked elements and continuity.
Use when analyzing script characters for the current Banana/Veo short-drama pipeline. Output a dual-layer character asset spec: a slim control_table for downstream automation and a lightweight production_bible for human review and reuse.
| name | voice-bound-shot-video |
| description | Use when generating a single talking-head or dialogue shot from a first frame, a minimal motion tag, a clean voice audio file, and dialogue text. This skill is for voice-locked shot rendering in the xiaolagumanju pipeline. |
把单个镜头的首帧图、极简动作提示、干净人声音频、对白文本，交给指定的 RunningHub AI 应用，生成一个音色和说话内容绑定的单镜头视频。
这个 skill 的定位是：
它不负责整片分镜选择，不负责全片拼接，也不负责自动决定哪个镜头该走这条链路。
适用于这些镜头：
不建议用于这些镜头：
输入建议通过一个 request json 统一描述，至少包含：
shot_idfirst_frame_pathvoice_audio_pathdialogue_textwidthheight同时必须提供下面二选一：
motion_tagmotion_prompt_cn可选字段：
character_idspeaker_nameproject_idnotes建议落盘这些文件：
shots/<shot_id>/input/request.jsonshots/<shot_id>/outputs/<shot_id>_voice_bound.mp4shots/<shot_id>/outputs/response.jsonshots/<shot_id>/manifests/voice_bound_result.json男人在说话、女人在说话 这一级别。走着说、边走边说 之类的动作要求时，才允许加入动作描述。dialogue_text 必须和音频语义一致，避免明显错字或不一致。motion_tag 和 motion_prompt_cn 冲突，以标准化后的 motion_prompt_cn 为准。第一版建议把 无动作说话 作为默认入口：
male_talk_neutralfemale_talk_neutralmale_talk_walk_slowfemale_talk_walk_slowmale_talk_emotional_lowfemale_talk_emotional_low推荐映射：
male_talk_neutral -> 男人在说话female_talk_neutral -> 女人在说话male_talk_walk_slow -> 男人一边缓慢走动一边说话female_talk_walk_slow -> 女人一边缓慢走动一边说话male_talk_emotional_low -> 男人在说话，情绪低压但克制female_talk_emotional_low -> 女人在说话，情绪低压但克制参考 request.schema.json。
最小请求示例：
{
"shot_id": "shot_001",
"first_frame_path": "shots/shot_001/input/first_frame.png",
"voice_audio_path": "shots/shot_001/input/voice_clean.wav",
"dialogue_text": "你终于回来了。",
"motion_tag": "male_talk_stand_calm",
"width": 736,
"height": 1280
}
参考 result.schema.json。
最小结果示例：
{
"shot_id": "shot_001",
"render_route": "voice_bound_shot_video",
"status": "success",
"input": {
"first_frame": "shots/shot_001/input/first_frame.png",
"voice_audio": "shots/shot_001/input/voice_clean.wav",
"dialogue_text": "你终于回来了。",
"motion_tag": "male_talk_stand_calm",
"motion_prompt_cn": "男人站着说话，情绪克制",
"width": 736,
"height": 1280
},
"output": {
"video": "shots/shot_001/outputs/shot_001_voice_bound.mp4"
},
"runtime": {
"provider": "runninghub",
"workflow_type": "voice_bound_video_app",
"app_url": "https://www.runninghub.cn/ai-detail/2036019863617015809"
}
}
motion_tag，先映射出标准 motion_prompt_cn。status: unsupported 或 status: failed，并写明原因。