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。
Motion-first AI video director for topic-to-video creation. Use when the user gives a theme, idea, article, script, or brief and wants AI to freely create a real animated video instead of a PPT/slideshow-style explainer; when deciding visual metaphors, motion grammar, scene beats, component choices, generated image needs, HyperFrames/Remotion implementation strategy, and anti-PPT QC.
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。
Act as the director layer that turns a topic into a motion-first video concept and controls execution until the result behaves like video, not a deck.
This skill does not replace ra-video-production-director, hyperframes,
remotion, image generation, or QC tools. It runs before and above them:
define the motion language, choose the visual system, reject PPT-like plans,
then hand off to the right production lane and keep the anti-PPT gate active.
Do not start from pages, slides, or cards. Start from movement.
For every topic, first find a visual metaphor that can move: a river, network, branching tree, orbit, factory line, swarm, collapse, compression, scanner, handoff, growth loop, map route, stack, wave, clock, lens, or machine.
Text is an anchor, not the main carrier. If a beat depends on reading a long paragraph, convert the idea into motion, structure, symbols, or images.
Capture only decisions that change the creative path:
If the user simply gives a topic and asks for a video, proceed with best judgment. Do not ask for a full brief unless the missing answer changes the production route.
For factual, historical, scientific, legal, medical, financial, current, or named-entity topics, verify the core claims before writing the motion plan. Prefer primary or authoritative sources. Use absolute dates when dates matter.
Keep facts short. The video should not become an encyclopedia; facts exist to support the motion narrative.
Write one sentence:
This video moves by showing <visual metaphor> transforming from <start state> to <end state>, proving <core claim>.
Examples:
If no moving metaphor is found, stop and invent one before planning scenes.
Plan a continuous timeline, not pages. Each beat must include:
time: start-end secondsnarrative job: hook, reveal, contrast, mechanism, consequence, proof, closemain moving object: the element carrying motionstate change: what physically changes on screencamera/layer motion: push, pan, parallax, orbit, crop, or stable basetext role: title, label, caption, counter, or noneasset need: code/SVG, generated image, screenshot, icon, footage, or nonePPT risk: what would make this beat feel like a slideAt least 80% of beats must have a visible state change beyond fade/slide-in.
Read references/motion-grammar.md before choosing primitives. Combine 2-4
motion primitives per video and reuse them with variation. Do not stack every
effect everywhere.
Use generated images only when they carry narrative work: a character, scene, object, historical visual, metaphor, or visual texture that code cannot express well enough. Avoid decorative stock-like images.
If image generation is used, specify:
Static images must be animated with crop, mask, parallax, reveal, light sweep, depth layers, or transformation. A still image plus text is not enough.
Default to HyperFrames for authored motion videos in this workspace. Read and follow the relevant HyperFrames skills before writing composition code.
Use Remotion when the project is already React/Remotion-based or when React component reuse is the main advantage. Use Lottie/Rive/Three.js only for the specific visual layer that benefits from them.
If the user asks for a complete production, continue into production instead of stopping at a plan:
<WORKSPACE>/01-内容生产/视频工作台/动效项目/<WORKSPACE>/legacy-archive/质检/ folderRead references/anti-ppt-gate.md before implementation and again before
final delivery. If the plan fails, rewrite the motion plan before coding.
Hard fail conditions:
Before calling a video done:
For planning-only requests, output:
For production requests, use the same structure internally, then produce the actual video and final QC artifacts.
references/motion-grammar.md: visual metaphors and motion primitives.references/anti-ppt-gate.md: checklist for rejecting slide-like videos.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。