batch-gen

Batch Video Generator

Generate multiple educational videos from a single JSON manifest. Reuses every script from edu-vid-gen — this skill only orchestrates the batch flow.

Scripts

All scripts at __PLUGIN_DIR__/scripts/. Batch-specific:

Script	Purpose
validate-manifest.py	Pre-flight schema check + cost estimate
batch-checkpoint.py	Read/write batch-status.json state tracking
checkpoint.py	Per-video phase gate (supports --json-output)
validate-sync.py	Subtitle overlay + Gemini 2.5 Pro sync analysis (Phase 5 self-healing)
composite-video-first.py	Video-first fallback compositor (Phase 5 attempt 3)

All generation scripts from edu-vid-gen are used as-is.

---

Phase 0 — Load & Validate Manifest

The user provides a batch.json file. For schema details: read references/batch-mode.md.

Step 0a — Validate manifest:

python3 __PLUGIN_DIR__/scripts/validate-manifest.py --manifest "{MANIFEST_PATH}"

If errors: show them and stop.

Step 0b — Present cost estimate. Wait for approval.

Step 0c — Initialize batch state:

python3 __PLUGIN_DIR__/scripts/batch-checkpoint.py \
  --init --manifest "{MANIFEST_PATH}" \
  --output-dir "{BATCH_OUTPUT_DIR}"

Step 0d — Create all video output directories.

---

Phase 1 — Derive Inputs (All Videos)

For each video, merge defaults + overrides to compute all variables.

python3 __PLUGIN_DIR__/scripts/batch-checkpoint.py \
  --update --video "{VIDEO_ID}" --state SCRIPTING --status-file "{STATUS_FILE}"

---

Phase 2 — Write Scripts (All Videos)

Follow edu-vid-gen Phase 2 rules exactly. After ALL scripts written:

BATCH REVIEW GATE: Scripts

Present all scripts. Wait for approval.

---

Phase 2.1 — Shared Characters (if any)

Generate character sheets for each entry in shared_characters. Wait 35s between calls.

BATCH REVIEW GATE: Characters

---

Phase 2.5 — Audio Timelines (All Videos, Sequential)

ELEVENLABS_API_KEY="$ELEVENLABS_API_KEY" node __PLUGIN_DIR__/scripts/generate-audio-timeline.mjs \
  --text "{FULL_NARRATION}" \
  --output-dir "{OUTPUT_DIR}/audio" \
  --voice "{VOICE_ID}" --model "{VOICE_MODEL}" \
  --stability 0.5 --speed 0.98 --language hi \
  --min-clip 5 --max-clip 8

After each: run rebuild-timeline.mjs and slice-audio.mjs.

BATCH REVIEW GATE: Timelines

After approval:

python3 __PLUGIN_DIR__/scripts/checkpoint.py --phase 2.5 --output-dir "{OUTPUT_DIR}"
python3 __PLUGIN_DIR__/scripts/batch-checkpoint.py \
  --complete-phase --video "{VIDEO_ID}" --phase "2.5" --status-file "{STATUS_FILE}"

---

Phase 2.7 — Strategy Selection (All Videos)

Choose ac_tc or extend from manifest or ask user.

---

Phase 3 — Generate Keyframe Images (Round-Robin)

Rate limit: 35s global between Gemini calls. Round-robin across videos.

After all images for a video: compress for Veo.

BATCH REVIEW GATE: Images

---

Phase 4 — Generate Video Clips (Sequential, Interleaved Validation)

Rate limit: 60-90s global between Veo calls.

For each clip: generate, validate with validate-clip.py, queue retries if score < 7.

BATCH REVIEW GATE: Flagged Clips Only

Auto-approve scores >= 7. Present only flagged clips.

---

Phase 5 — Self-Healing Composite (Parallel per Video)

Each video runs max 3 composite attempts:

1. Attempt 1-2: composite.py (VO-first)
2. Attempt 3: composite-video-first.py (video-first fallback)

After each: run validate-sync.py for Gemini Pro sync analysis.

If score >= 7: PASS. If < 7 after 3 attempts: ESCALATE.

---

Phase 6 — Summary Report

python3 __PLUGIN_DIR__/scripts/batch-checkpoint.py \
  --summary --status-file "{STATUS_FILE}"

---

Resume

/batch-gen --resume {path-to-batch-status.json}

---

Error Handling

Error	Strategy
One video fails	Mark FAILED, continue others
Gemini 429	Wait 60s, retry
Veo face-blocking	Switch to text-to-video
Auth errors	Pause entire batch
Self-healing exhausted	Escalate to user

---

Prompt Building

Load references as needed:

• edu-vid-gen/references/prompting.md
• edu-vid-gen/references/validation.md
• edu-vid-gen/references/audio-tags.md
• edu-vid-gen/references/transitions.md
• edu-vid-gen/references/api-errors.md