Photo to hand-drawn video

Read the applicable contract

• Read references/pipeline.md before preparing a target or running the planner.
• Read references/acceptance.md before accepting a preview or video.
• Read references/third-party.md before installing or redistributing runtime files.

Respect the hard boundaries

• Never composite the target image into a runtime frame.
• Never use

  drawImage

  , clipping masks,

  destination-in

  , image patches, or whole-canvas reveal strokes to imitate painting.
• Never optimize only for SSIM. Keep process authenticity, identity similarity, edge quality, and anti-copy checks separate.
• Never claim arbitrary-image coverage. This release supports one person or one centrally framed companion animal.
• Keep the original photo, generated target, music, models, and production video out of Git unless the user owns the rights and explicitly asks to publish them.

Prepare the target

1. Inspect the reference photo.
2. If the user did not supply a target, use an available image-generation tool to create one square stylized drawing.
3. Preserve identity, pose, crop, major clothing/fur markings, accessories, and primary scene objects.
4. Keep the target reproducible by marker, dry-brush, textured-ink, and liner strokes.
5. Sanitize image metadata with an available privacy skill before staging or publishing it.
6. Stop after three failed target attempts. Do not force manual coordinates.

Prepare dependencies

SKILL.md

SKILL_DIR

python3.11 -m venv .venv
.venv/bin/pip install -r "$SKILL_DIR/scripts/requirements.txt"
npm install
npm run install:browser
python "$SKILL_DIR/scripts/setup_runtime.py" --brushlib-dir /path/to/brushlib-wasm-build

ffmpeg

PATH

Run a preview plan

--run-dir

SKILL_DIR/scripts

.venv/bin/python "$SKILL_DIR/scripts/run_generalized_s.py" \
  --reference /absolute/path/reference.jpg \
  --target /absolute/path/target.png \
  --subject-type human \
  --run-dir "$SKILL_DIR/scripts/runs/example" \
  --budget-scale 0.5

--subject-type animal

auto

generalized-s-preview.png

semantic-preflight.json

Render the production video

--budget-scale 2.0

--music

.venv/bin/python "$SKILL_DIR/scripts/run_generalized_s.py" \
  --reference /absolute/path/reference.jpg \
  --target /absolute/path/target.png \
  --subject-type animal \
  --run-dir "$SKILL_DIR/scripts/runs/production" \
  --budget-scale 2.0 \
  --render \
  --duration 64.03 \
  --fps 25 \
  --music /absolute/path/music.mp3

--skip-video-validation

video-validation.json

Report the result

• detected contract and subject type;
• stroke count and budget scale from metrics;
• video duration, resolution, and validation result;
• any target-generation retry;
• whether music and privacy sanitization were applied;
• the current composition and visual-quality limitations.