What does adu-motion-video do?
The user provides a recorded voiceover video and script; ideally, they also provide a timecoded subtitle file for the new voiceover. When selecting scenes that include screen recordings, work walls, avatars, branding, or music, the user must also provide the assets they have the right to use for this episode. Your responsibility is to adapt the complete scene choreography to the new lines and runtime, generate an editable project, verify it, and export the new MP4. First, read the Production Conventions. SKILL_DIR is the directory of this file. PL="$SKILL_DIR/scripts/pipeline.sh". Production records save the actual Skill path, version, and entry point used; frozen execution code from an older installation with the same name, a repository version, and a previously generated project may differ—do not use the latest documentation to retroactively confirm an older project.
Choose Entry Point
When the user specifies a template ID (e.g., 01-B) or asks to choose a style, first read the Template & Style Catalog, mapping from "top-level template → secondary style" to an explicit package version. Unreleased styles cannot be promised as published packages. When auxiliary animation assets are needed, search and download them using the Public Animation Assets API into the current project directory, then verify the selected package's media and rendering requirements.
When the user adds a completed project, refines a new set of shots, or updates the template library, read Continuous Template Intake. First register the source and actual status, then refine based on value; receiving a project does not mean a publishable template exists. Use bash "$PL" packs only to list package semantic and version metadata, load relevant manifests based on the task; when multiple versions of the same ID exist, explicitly select id@version.
When the user specifies 03 · Dark 3D / 03-A · Multi-Screen Display, read the Dark 3D Template, and select [email protected]. It retains the nine-scene work wall, multi-screen comparison, zoom-in on key elements, and software window choreography, adding a native 1080×1920 portrait layout. Version 1.1.1 adds HDR frame-skipping and SDR export color verification; previous cold installation records belong to version 1.1.0, which still awaits the author's continuous audio-visual confirmation; select full scenes based on the current episode's materials, not counting them among the seventeen stable groups. The older version 1.0.0 remains unchanged.
- Default to running
bash "$PL" packs, select an explicit version based on the expressed purpose, read the selected package'sREADME.mdandmanifest.json, and review the Full Project Templating Guide.[email protected]retains staggered, reactive, and stabilizing movements across three validated groups;[email protected]retains object morphing, disassembly, and closure across two validated groups.[email protected]is one paper-perspective bridging group. Stability range is limited to validated macOS 1080p60, and does not imply the entire source or any script is templated; see the Capability Matrix for group boundaries. Old candidate versions are retained; planned interference groups for B remain candidates. Olderanim3/anim4have 12/9 full scenes respectively and are still experimental. Package IDs are unrelated to the original video title or brand. C/D/E's[email protected],[email protected]and[email protected]each have three stable groups, retaining front/back layer and character yielding, column-based and cross-column transitions, and whole-word and entity responses respectively. For each route, an executor not involved in the refinement independently produces a 99.5-second new video from the frozen public package and the current episode's input, confirmed valid by the author. Old candidate bytes are preserved; this episode's bridging does not count as a public stable group. - For lightweight short videos or layout exploration, use the 9 basic
TPL; they are quick-starts and do not include all actions from the full project packages. - For hand-drawn characters or continuation with matching shots, read the Hand-drawn Animation Method. There is currently no corresponding full project package; do not refer to the method documentation as automatically templated.
Version 3.3.0 adds 02-B · Paper Aggregation [email protected]: six types of materials aggregated into a network, plus two stable groups for a seven-step calendar, six capabilities, three unboxings, and a delivery card stack. Source/playback comparison and a 18.2667-second sample with different content have received the author's audio-visual confirmation. Requires the current episode's real voiceover/script and an explicitly provided title font with a fixed SHA; see the Package Description. This is a group-level increment, not a new full-video route; strict grid issues remain in review.json. The 3.2.0 fifteen stable groups and old candidates remain unchanged; total seven styles with seventeen stable groups.
Full Project Package Workflow
Version 3.6.0 adds current storyboard partial rematching and asset version checking, and integrates shared color processing for generic video frame-skipping. 04-A · Console Demonstration current candidate [email protected] adds bounded text layout; the original 0.1.1 three groups' same-content technical replay is retained; muted new script checking does not replace full audio-visual validation or independent reuse acceptance for a new voiceover; see the Template Description. New previews uniformly use the preview command with export evidence and version checking.
For a new script, first generate brief.json according to the Adaptive Composition System, then run bash "$PL" macro-plan packageID@version path/to/current-brief.json path/to/parent-dir/new-plan-dir. This entry filters by semantic phase, quantity, original-speed action windows, and preceding/following scene dependencies, then reduces effect repetition within valid combinations. 26 full groups across 8 style packages have code-verified adaptation profiles; capability is still adaptation-experimental. Outputs spec.json/report.json/report.md/profile.json; exit code 0 means binding check passed, 2 means draft saved awaiting completion or blocked, 1 means invalid input. If media or real keywords lack timecodes, modify brief and re-plan; do not fill in source demonstration timecodes. macro-spec is retained as a historical/manual recipe entry; it does not provide this layer of semantic and adjacency diagnostics; new productions should prioritize plans with adaptation, and the build will re-validate.
When a plan already exists and new assets are added, run macro-rematch according to Local Storyboard Rematching, review suggestions, then save to a new directory via macro-apply-rematch. Fix original segment semantics, timing, and other storyboards, re-validate media and bilateral seams; reject application if baseline or assets have changed. If no compatible new group exists, retain original groups; do not randomly swap effects or guess script mappings for other templates. This interface is intended for production tool calls and does not yet equal integration into the ADuDir drag-and-drop interface.
- Run
bash "$PL" doctorto check Node, Python, FFmpeg, and Chrome; if dependencies are missing and installation is permitted, then runbash "$PL" setup. Keep the source project and original output file read-only, and output to a new path that doesn't exist. - First, lock the user-specified template source. Other
animfiles, new projects, or old outputs in the assets directory are not automatically animation blueprints; do not claim the entire video uses the specified template just by inserting a few template fragments. Record the source package and scene ID for each scene; for mixed full packages, see Sources and Multi-Package Orchestration. Align with the new voiceover and script to obtain accurate timings for key lines. When SRT only has whole-sentence timings, do not guess individual word timings within a sentence; use word-level alignment data or manual annotation. Read the package'srole,cues,motionWindows, andslots; select, reorder, or repeat full scenes based on meaning. First list the storyboards and missing assets. - Write the brief based on the new voiceover, run
bash "$PL" macro-plan packageID@version /path/to/current-brief.json /path/to/existing-parent/new-plan-dir; review the report, fill in missing items in the brief if necessary and re-plan, then build using the generatedspec.json. Before rendering, verify phases against actual lines, key event positions, asset identities, and whole-video repetition according to Reuse Content Verification. The planner checks the declared structure; it does not judge whether the filled phases are faithful to the lines. Usemacro-speconly for manually maintained historical recipes; its cues are left empty, and example text and original video duration are only for understanding the interface; you must fill in the current episode's content and actual timecodes; do not mechanically add the same offset to every repeated group.brandshould be the current episode's account. Music can be specified withmusic:{mode:"track",path,offset}; if unspecified, the original composition structure will be re-synthesized. By default, the character window uses Vision to track the new voiceover's face on macOS; if automatic tracking is not possible, explicitly configure a reviewedfaceTracking:{mode:"fixed",cx,cy,h}. Any source demonstration data and example text must be replaced or clearly marked as "schematic". An empty path for an asset represents a missing item; media preparation and capacity limits are in the package directory'sREADME.md. - Run
bash "$PL" macro-build packageID@version /path/to/new-plan/spec.json /path/to/edited-voiceover.mp4 /path/to/new-project [--subs-js /path/to/subtitles.js]. The new voiceover duration must be roughly consistent with the planned total frame count. The adapter locks key action windows to original speed, extending only holdable areas for longer passages; if time is insufficient, it will error, requiring a change of storyboard or segment length—do not rely on full-video linear stretching. Missing text, numbers, media, or key cues will also cause errors. - Run
bash "$PL" stills /path/to/new-project 0,4,12,...to spot-check the first frame, before/middle/after each cue, longest text, scene seams, and end card; small circular windows must show both eyes, mouth, and chin; fixed parameters must be determined from the current episode's assets and check extreme poses like looking down or up; then play all complex motion segments, comparing against the original project. Check fonts, media aspect ratio/cropping, actual lip-sync, subtitles, SFX, and music. Revise configuration or project and re-validate. - Run
bash "$PL" audio /path/to/new-project→bash "$PL" mix /path/to/new-project→bash "$PL" render /path/to/new-project /path/to/new-video_v1.mp4→bash "$PL" check /path/to/new-video_v1.mp4. Watch and listen to the final encoded file in full; frame counts, decoding, and loudness fromcheckcannot replace visual and content acceptance. New previews are uniformly published viabash "$PL" preview /path/to/new-project /path/to/new-video_v1.mp4 /existing-parent/new-preview-dir, binding the real output SHA, displaying the playback version and prompting old tab updates. HDR/DV, source bit depth, and metadata risks must be documented with actual comparison evidence according to Asset Color and Preview, not just label checks.
For a quick experience with basic templates, use bash "$PL" demo /non-existent-new-directory. When composing TPLs from scratch for a real voiceover, follow the Template Directory and Subtitle Options, and still perform full validation.
Non-Negotiable Checks
- Voiceover frame grabs must use the target output clock; source scene animations run on the mapped original choreography clock. Do not let the original scene clock control real lip-sync, and do not use frozen real-person frames to compensate for missing duration.
- If the first frame is the voiceover opening, the real person and topic should be visible; check subtitle white text stroke, font size, and safe area; do not leave the template source video's branding, dates, numbers, or placeholder text in the new video.
- Check continuous segments within every action protection window, not just a single screenshot; verify screen recordings, nine-grid, and work walls against the current episode's real assets. Music and sound effects are re-composed along the new timeline; do not simply speed-change old output audio tracks.
- A new script does not guarantee automatic lossless adaptation. If a short segment cannot accommodate the actions, reject, split, or change the scene; long segments can only be extended at allowed hold positions. New scenes still need to be designed for the current episode's content; do not forcibly fit sentences into macro-scenes.
- Output the new project and the new version of the output file; do not overwrite the user's original project. Report the actual scope of the check.
anim3andanim4remain experimental until full frame-by-frame comparison and audio-visual acceptance of the output for a complete new script; do not claim they are fully validated.
More details: Full Project Templating · Production Conventions · Subtitles · Audio · Hand-drawn · Troubleshooting.