P(doom)! Kaboom! #

Input assets and source for the latest animated music video: the orange character in a straw boater and bowtie, a clone chorus, changing stage and isolated shots, bottom lyrics, and a two-state mouth driven by the vocals.

Contents #

• assets/audio/P(doom)! Kaboom!.m4a — the original song, copied without re-encoding.
• assets/character/ — all three original character references, copied without alteration.
• assets/lyrics/lyrics.txt — lyrics extracted from the song's embedded metadata.
• assets/lyrics/source-lyrics.srt — the original embedded lyric timing used for alignment.
• assets/manifest.json — SHA-256 checksums and sizes of the input assets.
• src/create_video_v3.py — current choreography, camera cuts, poses, mouth animation, and captions.
• src/create_video_v2.py and src/create_video.py — shared geometry and drawing helpers retained from earlier versions. Use render.py to render the current version.
• data/ — saved vocal alignment, 100 Hz mouth timing, lyric cues, shot list, chapters, and subtitles. Rendering does not require downloading any models.
• src/prepare_vocals.py, src/align_vocals.py, and src/build_mouth_track.py — optional scripts for rebuilding the vocal timing.
• docs/original-v3-validation.json — validation measurements from the previously delivered full video.

Rendered movies, decoded audio, virtual environments, and model caches are excluded from Git. The reference images guide the character design; the renderer draws its geometry procedurally and does not load the images into each frame.

Render #

The tested setup uses Python 3.12, FFmpeg on PATH, and macOS supplemental fonts: Trebuchet MS Bold, Rockwell, Copperplate, and Baskerville. Fonts are not bundled. Set PDOOM_FONT_DIR to another directory containing these font files if needed.

python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt

# A single frame, then a short preview with the song.
python render.py --still 3.05
python render.py --start 1 --end 6

# Full 1920×1080 video, 30 fps, 151.6 seconds, with audio and burned-in lyrics.
python render.py

Results are written to renders/; scratch files go to work/. Full rendering is CPU-intensive. Run python src/v3_scene_sheet.py to rebuild the scene overview and poster. These commands can be run from any directory if the script paths are adjusted; asset paths are resolved relative to the repository.

Edit SHOTS and Performer in src/create_video_v3.py to change the performance. The renderer reads the saved captions from data/v3-cues.json and the binary mouth track from data/v3-mouth-track.npz. The shot list JSON is a reference export; the Python SHOTS list controls the edit.

Optional: regenerate the vocal timing #

The saved timing already captures the latest edit. To recompute it, use a separate Python 3.12 environment. The first run downloads Demucs and Wav2Vec2 model weights; processing then runs locally on the CPU. Model weights and extracted stems stay in the ignored work/ directory.

python3.12 -m venv .venv-audio
source .venv-audio/bin/activate
python -m pip install -r requirements-audio.txt
python src/prepare_vocals.py
python src/align_vocals.py
python src/build_mouth_track.py

# Adopt the newly generated timing for subsequent renders.
cp work/v3-aligned-characters.json work/v3-aligned-words.json data/
cp work/v3-mouth-track.npz work/v3-mouth-open-intervals.json work/v3-cues.json data/

# Return to the rendering environment.
source .venv/bin/activate
python render.py --start 1 --end 6

Review changed captions and transitions before rendering the full video. The existing chapter metadata is fixed to this song. Rebuilt subtitles are written alongside the full render; data/P-doom-Kaboom-v3.srt preserves the delivered edit.

Asset provenance #

The audio and three character images were supplied for this project. Renaming the reference files did not change their contents. The lyrics text is extracted from the supplied audio; the original user wording is retained in the lyric cues. The source and timing data were created while making the video. No redistribution license is inferred for the supplied assets, and no third-party fonts or model weights are included.