AI剪口播
無料口播视频转录和口误识别。生成审查稿和删除任务清单。触发词:剪口播、处理视频、识别口误
日本語の概要は準備中です。原文の説明を表示しています。
End-to-end Chinese video washing pipeline. Use when the user provides a Bilibili, Douyin, Xiaohongshu, YouTube, web video URL, or local video file and asks for 视频洗稿, 视频二创, 洗稿并制作视频, 链接视频改写, 提取逐字稿后洗稿, or producing a new AI video from a source video. Orchestrates ra-逐字稿提取skill and ra-洗稿, ending at the 01-内容生产/视频工作台/待制作/ handoff queue; production continues through ra-video-production-director only on explicit request (直接制作/直接出片/一条龙).
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Turn a source video link or local video file into a production-ready washed script: extract the faithful transcript, rewrite it through the established ra-洗稿 stack, then queue the approved script into 01-内容生产/视频工作台/待制作/ as the binding production contract. Video production is a separate stage picked up from that queue (normally by the production host through ra-video-production-director); continue into production in the same session only when the user explicitly asks.
This skill is an orchestrator. Do not duplicate the detailed rules from ra-逐字稿提取skill, ra-洗稿, ra-人话, dbs-*, or ra-video-production-director; load and follow those skills at the step where they own the work.
When this skill is used for video washing, the source transcript is an internal working artifact, not a deliverable.
Hard rules:
Use this skill when the input is a video URL or local video file and the user wants a rewritten or newly produced video.
Do not use this skill for narrower tasks:
ra-逐字稿提取skillra-洗稿ra-video-production-directorIf the request says 视频洗稿, 二创成新视频, 洗稿并制作视频, or similar, assume extract -> rewrite -> handoff queue. Continue into same-session production only when the user explicitly says 直接制作, 直接出片, or 一条龙. Narrow the scope when the user explicitly asks for 只要逐字稿, 只要脚本, 先不要生成视频, or 只分析.
Capture the minimum decisions:
16:9 landscape video when unspecifiedra-video-production-directorAsk only when the answer changes the execution path. Otherwise choose the conservative default and proceed.
查重硬门:拿到源链接后、提取逐字稿前,先运行 python3 automation/scripts/wash_ledger.py check "<url>"。退出码 3 = 已洗过:停下,把台账记录(时间、选题卡、产出去向)报给用户,用户明确确认重洗才继续;退出码 4 = 曾标记不做或标题疑似:提示一句后继续。查重结果只在会话里说,不写进任何产物。
Invoke ra-逐字稿提取skill first, but use it as an internal dependency.
For Douyin/Xiaohongshu URLs, prefer:
python3 "$ZT_HOME/scripts/transcript.py" "<url>" --internal --output-dir "<temporary_internal_dir>"
Required outcomes:
If extraction fails because the platform blocks access or the link is private, stop and ask for a local file, public link, or exported captions. Do not continue from a guessed transcript.
Pass the complete transcript and user goal into ra-洗稿.
Require ra-洗稿 to run its default stack:
ra-人话 rewritedbs-ai-check diagnostic passdbs-hook first-five-seconds passdbs-resonate resonance passra-video-title candidates pass: 8-12 candidates with a Top 3 recommendation into the handoff 标题候选 section (no final title — the user picks manually at publish time)dbs-xhs-title only when Xiaohongshu 图文/笔记 titles are in scopeExpected rewrite output:
ra-video-title)If the user asks for 只要脚本, stop here and return the rewritten script and flow checks only. Do not return the source transcript path.
Require ra-洗稿 to finish with its default endpoint: a handoff file in
01-内容生产/视频工作台/待制作/YYYY-MM-DD-<主题slug>/交接稿.md following 01-内容生产/视频工作台/_交接模板.md (create the project folder first; the handoff lives inside).
Before stopping, verify:
status: 待制作, platform, ratio defaulting to 16:9, duration_target_s,
word_count, voice: pluvio-indextts2-calm-v1,
voice_provider: indextts2-local,
captions: asr-word-timestamps, caption_style: anchor-dark, visual,
illustration_skill)ra-video-title; no final title is picked — the user selects manually later台账登记(硬门):交接稿与选题卡落地后,登记洗稿台账——优先 python3 automation/scripts/wash_ledger.py add --status washed --from-source-json "<.internal/<日期>-<平台>-download/…source.json>" --card "<选题卡相对路径>" --handoff "<交接稿相对路径>"(本流水线下载源视频时一定有这个文件);未走下载、只有链接时用 --url "<源URL>" --title "<源标题>";来自云端选题推荐时加 --work-id <云端作品id>。用户看完源决定不洗时,也要 add --status skipped --url "<源URL>" --note "<原因>",防止之后重复推荐。
Then stop and return the handoff file path plus the script and flow checks. Production is picked up from the queue as a separate stage.
Only when the user explicitly said 直接制作, 直接出片, or 一条龙:
invoke ra-video-production-director with the handoff file as the binding
production contract, plus:
ra-洗稿 flow-check results16:9 landscape canvas with the white-first Xiaohei visual system unless the user gave a different styleian-xiaohei-illustrations whenever the user
asked for 小黑配图 or the white-first Xiaohei default is selectedFor a first run, changed style, or uncertain motion direction, render a 10-second sample first. Continue to the full video only after approval unless the user has explicitly asked to skip the sample gate.
Let ra-video-production-director run its normal QC and archive rules.
Minimum final checks:
01-内容生产/视频工作台/已制作/<月上旬|月下旬>/<日期-主题>/成片/automation/scripts/check_delivery.py passes against the handoff file, project dir, and final MP4status: 已完成, and 01-内容生产/视频工作台/已制作/<月上旬|月下旬>/<日期-主题>/ is assembled with the handoff file, the final MP4 copy under 成片/, and key QC frames under 质检/ra-audio-to-subtitles; caption-qc.json passed; skill-captions applied
the handoff caption_style (default anchor-dark) and produced a passing
caption-render-qc.json; character-count timing was not usedian-xiaohei-illustrations,
saved as project assets, loaded directly by the render, and explain the
current point instead of acting as decorationFor the default handoff completion, return:
交接文件:
[01-内容生产/视频工作台/待制作/YYYY-MM-DD-<slug>/交接稿.md]
标题候选:
[Top 3 推荐;完整候选见交接文件]
新逐字稿/脚本:
[script]
流程检查:
- AI 味:
- 开头:
- 共鸣:
- 小红书标题(如适用):
For script-only completion (只要脚本), return the same block without the
交接文件 line and do not create a queue file.
For explicit in-session production completion, return:
成片:
[final MP4 path]
质检:
[key QC path]
过程资产:
- 新脚本:
规格:
[ratio, resolution, duration, codecs]
限制:
[only if something remains]
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
口播视频转录和口误识别。生成审查稿和删除任务清单。触发词:剪口播、处理视频、识别口误
日本語の概要は準備中です。原文の説明を表示しています。
将法天象地、同源巨大法相或角色力量显现的视觉参考制作成连续关键帧,再使用用户选择或可用的图生视频工具分段生成、检查和合成动作视频,不绑定成片平台。适用于本体与法相同框、领域展开、凝实与同步出招;不负责整部小说改编或普通图片轮播。
日本語の概要は準備中です。原文の説明を表示しています。
口播基础素材包生成。转录口播、识别口误、生成审核页;用户确认后剪出新视频,Agent 再基于剪后视频重新转写、AI 校对字幕,输出后续口播成片可用的 source_cut.mp4 和 subtitles.srt。触发词:剪口播、处理口播素材、准备口播素材、识别口误、基础素材包
日本語の概要は準備中です。原文の説明を表示しています。
口播视频成片 Skill。把文章/口播稿/SRT、剪后视频和 HTML/图片素材串成分镜稿、时间线预览和最终 MP4;成片比例和动画风格从用户配置读取,动画默认使用小黑风格。触发词:口播成片、做分镜稿、时间线预览、合成口播视频、导出竖屏MP4
日本語の概要は準備中です。原文の説明を表示しています。
自进化 skills。记录用户反馈,更新方法论和规则。触发词:更新规则、记录反馈、改进skill
日本語の概要は準備中です。原文の説明を表示しています。
dontbesilent 商业工具箱主入口。双模式:任务前路由(你的问题该用哪个 skill)+ 任务后导航(刚做完诊断,下一步该干什么)。 触发方式:/dbs、/商业、「帮我看看」、「下一步怎么走」 Main entry point for dontbesilent business toolkit. Dual mode: pre-task routing + post-task navigation. Trigger: /dbs, "help me with my business", "what's next"
日本語の概要は準備中です。原文の説明を表示しています。