hand-drawn-infographic-video-board

Hand-Drawn Infographic Video Board

Core Rule

Do not treat a generated PNG as the animation source of truth. Use a dual-layer package:

visual layer: board.png, svg_preview, or url from board_asset_manifest.json
control layer: board_manifest.json + annotation_manifest.json + motion_plan.json + optional board.svg/board.html

The video may show board.png, but cursor paths, underlines, circles, boxes, checks, strike-through marks, and camera targets must come from explicit canvas-pixel coordinates in the control layer. Never use vague regions such as "upper left area" or "around the title".

Workflow

1. Read references/contracts.md before changing package shape or producing artifacts.
2. Prepare a compact board_spec.json. Keep only key visual objects, and give important objects stable ids.
3. Generate or provide board assets through board_asset_manifest.json. For local smoke tests, prefer asset.kind=file or asset.kind=svg_preview.
4. Provide voiceover_segments.json with segment ids, text/caption, targets, board ids, and action-level spokenAnchor values.
5. For legacy single-board packages, run:

python3 scripts/generate_board_package.py \
  --input path/to/board_spec.json \
  --board-image path/to/board.png \
  --voiceover path/to/voiceover_segments.json \
  --output path/to/board-package

1. For project/multi-board packages, run:

python3 scripts/generate_board_package.py \
  --project path/to/project-output \
  --asset-manifest path/to/project-output/board_asset_manifest.json \
  --voiceover path/to/project-output/script/voiceover_segments.json \
  --calibration-dir path/to/project-output/calibration \
  --output path/to/project-output/board

1. If an AI-generated PNG drifted from the control layout, run auto-calibration first. It detects element bboxes from the actual PNG and writes calibration/<boardId>.element_bboxes.json:

python3 scripts/auto_calibrate.py \
  --project-dir path/to/project-output \
  --provider auto \
  --write-tool-on-partial \
  --json