huodebing-alt
UserMulti-agent template for automated headshot/talking-head explainer video production. Script -> voice clone -> AI avatar -> rough cut -> B-roll -> final render. Model-agnostic, plug-and-play with /onboard.
Categories
Indexed Skills (50)
add-bilingual-subtitles
Produce a bilingual subtitle file with stacked Chinese (top) and English (bottom) lines per entry. Use for bilingual explainers and cross-market releases.
add-subtitles-burned
Burn styled subtitles into a video using ffmpeg's libass with a brand-aware style file. Use for platforms or audiences that won't toggle CC.
add-subtitles-srt
Emit a clean SRT sidecar from the voice track and segments index, suitable for YouTube CC upload. Use when subtitles should remain toggleable.
bilingual-explainer-template
Project template that produces both a Chinese and an English version of the same explainer with synced segmentation and bilingual subtitles. Use for cross-market releases.
broll-only
Render only the B-roll plan against an existing script; produces all B-roll clips without touching A-roll or audio. Use to iterate on visuals independently.
broll-with-manim
Generate math, geometry, or programming-flavored B-roll using Manim as a backup to Hyperframes templates. Use when topics require precise mathematical animation.
build-timeline
Build a master timeline JSON merging A-roll, B-roll, narration, BGM, and subtitles with absolute timestamps. Use as the source of truth for render-final.
check-visual-consistency
Verify brand palette, font, lower-third style, and avatar framing remain consistent across all segments and B-roll clips. Use late in QC.
clone-voice
Clone a user's voice via ElevenLabs Instant Voice Cloning from a short audio sample. Use once per project when the creator wants their own voice rather than a stock voice.
cut-fillers
Apply a cut list (from detect-fillers) to a video using ffmpeg, producing a tighter edit. Use after the user reviews and approves the cut list.
demo-video-template
Project template tuned for product demos — screen-cap intercut, fast pacing, headline emphasis. Use when the deliverable is a product walkthrough.
design-broll
Plan B-roll cues for a script — choose type (animation, data viz, flowchart, concept diagram, transition, subtitle emphasis) and timing for each cue. Use after script is finalized and before B-roll rendering.
detect-fillers
Scan an existing transcript JSON for fillers and output a cut list with start/end timestamps. Use to inspect what would be cut before actually trimming the media.
detect-loudness-issues
Measure LUFS, true peak, and loudness range for the final audio and flag platform-spec violations. Use right before encoding.
export-bilibili
Re-package the master to Bilibili's recommended spec — 1080p H.264 high profile, AAC stereo, bitrate cap aware. Use right before B站 upload.
export-tiktok
Export a 9:16 short cut (≤ 60s) optimized for TikTok with safe-area-aware crop and burned subtitles. Use to repurpose long-form into a TikTok clip.
export-wechat-channels
Export both vertical 9:16 and horizontal 16:9 cuts sized for WeChat Channels (视频号) limits. Use to publish to 视频号 with safe crops.
export-youtube
Re-package the master mezzanine to YouTube's recommended spec — 1080p H.264 yuv420p, AAC stereo, moov atom at front. Use right before uploading.
extract-avatar-headshot
Crop a clean headshot from a user-supplied photo or video frame for HeyGen avatar enrollment. Use during onboarding to prepare the avatar source image.
extract-key-terms
Extract a glossary of jargon, product names, and proper nouns from a script for consistent pronunciation and subtitle spelling. Use before TTS so the voice model receives a pronunciation lexicon.
fetch-asset-from-gdrive
Download referenced assets (footage, BGM, brand kits) from a Google Drive folder using shareable links or the Drive API. Use during setup to pull project assets onto disk.
full-pipeline
Run the entire video production pipeline end-to-end — Phase 1 script through Phase 5 export — with checkpointing. Use to go from brief to publishable file in one command.
generate-avatar-segment
Render a single 45–60s HeyGen avatar video from a segment audio file and avatar id. Use to produce one A-roll segment; the per-segment ceiling is enforced.
generate-script
Generate a full explainer-video script from a topic brief, target audience, and duration. Use when starting a new video project from scratch or rewriting an outline into a shootable script.
generate-thumbnail
Generate a 1280x720 thumbnail from a high-contrast frame plus a headline overlay. Use to produce YouTube/B站 thumbnails consistently.
headshot-only
Run only the A-roll path — script → TTS → HeyGen avatar segments → stitch — without B-roll or final assembly. Use to preview the avatar performance before investing in B-roll.
heygen-webhook-listen
Run a short-lived local webhook receiver for HeyGen async render callbacks during a batch, updating segment manifest as videos complete. Use when polling is undesirable or rate-limited.
lecture-video-template
Project template tuned for academic-style lectures — measured pacing, diagram-heavy B-roll, slide-like transitions. Use when the deliverable is a course module.
mark-chapters
Generate YouTube chapter markers from segment titles and section headers. Use to improve discoverability and viewer navigation in long-form videos.
mix-audio
Mix narration and BGM with sidechain ducking, loudness normalization to broadcast targets, and per-segment level matching. Use to produce the master audio bed for final render.
mux-voice-track
Concatenate per-segment narration MP3s into a single audio track with deterministic gaps between segments. Use to produce the master voice track for downstream timeline assembly.
regenerate-failed-segment
Re-render avatar segments flagged as failed by validate-lip-sync or by manual review, with one parameter perturbation per attempt. Use to recover specific bad segments without redoing the batch.
render-broll-animation
Render a single animated B-roll clip from a prompt using Hyperframes (HeyGen's open-source HTML→MP4 renderer). Use for general motion-graphics cues.
render-concept-diagram
Render an animated conceptual diagram (boxes, arrows, labels) from a structured spec using Hyperframes. Use when a flowchart is too rigid and you need a free-form illustration.
render-data-viz
Generate an animated chart B-roll from a CSV using a Hyperframes template. Use when a script cue calls for numbers, comparisons, or trends.
render-final
Render the final video by encoding the timeline to a master mezzanine MP4 using ffmpeg. Use as the last step before per-platform exports.
render-flowchart
Render an animated flowchart B-roll from a Mermaid/DOT graph using Hyperframes. Use for process explanations, decision trees, or system diagrams.
render-subtitle-emphasis
Render a kinetic-typography emphasis B-roll that highlights a single phrase from the script. Use to drive a key point home visually.
render-transition
Render a short transition B-roll (wipe, zoom, glitch, brand sting) between two sections. Use to bridge topic shifts in the final timeline.
review-pacing
Score final video pacing — shot length distribution, B-roll density, attention sag points — and recommend tightening. Use post-edit before final encoding.
review-script-pacing
Audit a script for pacing problems — overlong sentences, missing pauses, hook quality, and section balance. Use after generate-script and before segmentation to catch readability issues early.
rough-cut-pipeline
Run the full user-headshot rough-cut chain — transcribe → detect fillers → trim pauses → cut fillers — as one command. Use to go from raw recording to clean rough cut.
segment-script-45s
Split a full script into HeyGen-compatible 45–60 second segments at natural sentence boundaries. Use when preparing avatar A-roll rendering, since HeyGen has a hard per-segment ceiling.
stitch-avatar-segments
Concatenate validated A-roll segments into a continuous A-roll track with optional crossfade. Use after all segments pass QC to produce the master A-roll for editing.
test-voice-sample
Render a short test phrase with the configured voice to verify identity, pronunciation, and parameter quality before batch TTS. Use after clone-voice or after tune-voice-params.
transcribe-with-disfluencies
Transcribe a user-recorded video via AssemblyAI with disfluency tagging (um, uh, like, repeats, long pauses). Use as the first step of rough-cut when the user films themselves.
translate-script
Translate a script between Chinese and English while preserving section structure, BROLL cues, and pacing markers. Use when producing a bilingual explainer or localizing an existing video.
trim-pauses
Compress silences longer than a threshold in a video to a target length, preserving natural rhythm. Use to tighten user recordings without changing pacing.
tts-batch
Render TTS audio for every segment listed in segments.index.json with budget guard, retries, and a manifest. Use after segmentation to produce the full narration track set.
rough-cut-only
Take a user-recorded headshot video and produce a clean rough cut (transcript + filler removal + pause trim). Use when the user wants to use their own footage instead of an avatar.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.