Vibe Knowledge Awards Production
Turn a knowledge topic into a viewable short video and publishing assets. Supports full production from scratch, as well as modifying only subtitles, pacing, music, cover, or copy. When making partial modifications, preserve already-confirmed content and deliver only the parts requested by the user.
Confirmed Default Preferences
Default values are in assets/defaults.json. User requests and existing project settings take precedence:
- Chinese knowledge content; landscape 16:9, 1920×1080, 30fps; typically 30–45 seconds, determined by content and music beat, not forced to a fixed duration.
- No voiceover or dubbing. Keep original vocals from the user's music; "no extra dubbing" does not mean pure instrumental.
- Large text, short sentences, fast transitions, clear visual focus. Dark background + mint green + warm orange is a usable starting point; do not lock all subjects into a circuit-board style.
- Use the user's provided music, keep its original tempo, analyze beat points before arranging cuts. 4/6/8 beat pattern alternation is allowed, but adjust based on the actual rhythm of the track and reading time.
- Full delivery includes MP4, 4:3 horizontal cover, 3:4 vertical cover, title, description, hashtags, and source code for further modification.
- Default title format:
Vibe Knowledge Award|Specific Hook or Question, then add the knowledge name as needed. - Output is placed in an independent project within the current workspace; provide clickable absolute file paths at the end.
Taking Over a Task
First, check the current project for existing project.json, storyboard, preview, final video, and user-confirmed opinions. Do not redo already completed parts. When topic, music, specifications, or authorization are already clear in the conversation, do not ask again. Only ask when missing input truly affects the work, and first advance parts that do not depend on the answer.
Minimum input is the topic; music can be added later. If there is no music, first write the copy and storyboard; do not fabricate a music path or claim beat-matching is complete. When the user says "use your own judgment," use default preferences to autonomously complete standard choices. Preview confirmation and authorization for local export remain valid; authorized rendering, fixes, and retries do not need repeated confirmation. If the user only wants a preview, stop at the preview stage. If export is not yet authorized, provide a reviewable version and confirm once only. This process does not include automatic platform publishing.
Production Workflow
- Set up the project. Use
scripts/create_project.pyto create an independent project, saving topic/music/specifications. If a project already exists, continue directly. The script only creates the project directory and input record; it does not replace creative work, beat analysis, or rendering. - Verify the knowledge. Prioritize primary sources; clarify core conclusions, applicable prerequisites, one verifiable calculation example, and easily misunderstood boundaries. Save to
sources.md. Mark metaphors clearly; do not rewrite a physics theorem as a psychology law. Ensure consistency in mathematics, circuits, units, and signs. - Write short copy and storyboard. Start with a specific question; each shot carries one key point; end by returning to the question. Title, description, and graphics each bear different information; avoid having multiple explanations on one screen. First determine reading load, then arrange fast/slow pacing. Do not substitute shot count for impact.
- Music and unified timeline. When beat-matching is needed, read references/timing-and-render.md. Use
scripts/beat_plan.py analyzeto measure beats, checking BPM, half-beat/double-speed ambiguities, and strong onsets. Use the user's selected segment or trial listening result. Useplanto convert events such as shot beats, keywords, formulas, and corrections into a unified frame number and timeline. Music cut points, scenes, subtitles, and motion effects all use this table; do not maintain multiple sets of manually entered seconds simultaneously. - Build the project and preview. Default priority is locally available HyperFrames project. When needed, read available
hyperframes-vibe-169and its composition contracts, complying with sub-composition and paused GSAP timeline requirements. Use assets/hyperframes-clock.js or equivalent precision correction to avoid GSAP's six decimal places causing a one-frame delay in cuts. When switching engines, reuse the user's existing project; do not redo it just to call the skill. - Check and complete rendering. Run engine validation, check each shot, large text boundaries, random playback jumps, frames before and after key beats, then export locally based on existing authorization. Use
scripts/audit_render.pyto check the actual output video for specifications, frame count, complete decoding, and audio track offset. Additionally, inspect every shot transition and key motion effect frame by frame; script decode passing does not replace visual inspection. After fixing any issues, re-run only affected checks and final output verification. See timing reference for details. - Covers and publishing assets. Read references/visual-and-delivery.md. Use imagegen or existing clear cover templates to generate 4:3 and 3:4 independent compositions; verify Chinese characters, formulas, arrows, margins, and actual pixel ratios. Generate title, description, and a few relevant hashtags consistent with the final video.
- Delivery. After confirming the actual files exist, provide MP4, dual covers, publishing copy, and project directory. State duration, specifications, and important limitations only. If export was unsuccessful, clearly say "preview/source code is complete," and do not claim the final video is done. Keep project.json, sources.md, storyboard.json, beat-map.json, validation results, and source code for subsequent modifications.
Combining with Other Skills
- hyperframes-vibe-169: When choosing this engine, it handles composition structure, lint, preview, render. Its style examples do not replace the current subject.
- lemo-opuscar: Use when the user requests this skill or the subject needs its direction/style method; just read the corresponding style. For partial revisions to an existing final video, do not automatically re-run the new video's style questionnaire.
- painted-animation / Joker-chichi/claude-mv-skill: Only read relevant parts when hand-drawn animation, lyric MV, or its beat analysis tool is needed. Knowledge videos with BGM do not require the user to provide an LRC file.
- imagegen: Responsible for generating or modifying bitmap covers. Save final images to the project; record prompts. Do not treat the skill description as a directly callable image API.
- openfilm / jianying-editor: Only use when the user specifies or an existing project actually uses them.
Not all skills need to run simultaneously for each episode. When tools are unavailable, first check the local runtime environment and feasible alternatives; do not silently install large dependencies or replace user-specified products. Keep clear failure messages; do not mask issues with infinite retries.
Helper Scripts
Scripts run with Python 3. Audio analysis and sync checks additionally require NumPy; FFmpeg/FFprobe can be passed from PATH or explicit arguments. No API keys needed; no accompanying music, renderer, or third-party binaries. Use Python/Node found on the machine; do not treat old project absolute paths as universal dependencies.
python scripts/create_project.py --workspace <workspace> --slug <short topic name> --topic <topic> --music <MP3 optional>
python scripts/beat_plan.py analyze --music <music> --out <beat JSON> --offset 0 --seconds 50 --ffmpeg <ffmpeg path>
python scripts/beat_plan.py plan --grid <beat JSON> --story <storyboard JSON> --out <beat-map.json>
python scripts/audit_render.py --video <MP4> --plan <beat-map.json> --score <final music file optional> --out <audit.json>
Storyboard input format, manual beat correction, and render precision are detailed in the timing reference. Directory template, cover, and copy acceptance criteria are detailed in the visual and delivery reference. Script output values are evidence for checking, not automatic guarantees of artistic or scientific correctness.