The person brings a song (and its lyric sheet if they have one), the idea for the
video and the look they want. You deliver a 16:9 1080p video where drawn characters act
the song out, every cut lands on the song's bar grid, and kinetic captions hit the sung
syllables and ride the beat. Nobody sings on camera; for lip sync and live-action
scenes, workflow-multi-style-music-video is the route.
Ask at kickoff: the aspect, whether characters lip-sync (no, here), and who sings when
you can't tell from the song. The song's timing is guide-time-a-video-to-a-song
(bar-grid route); the look and its stills are guide-pick-a-stylized-look. Rounds and
deliverables follow the core skill.
Steps
- Map the song (
guide-time-a-video-to-a-song, Process steps 1–4): stems, sung words, the bar grid printout and the timed lyric sheet. Under $1. - Treatment (a prose file in the job folder): the concept, a beat map per section (bar and time, picture, caption idea), and the two caption layers: lyrics baked into the picture (carved, painted, written on things) and kinetic type over the cut. Gate: the person agrees it before anything is generated.
- Style frames (
guide-pick-a-stylized-look, Process step 1): four options per look, with a lyric baked into each frame to judge how type reads in that look. Gate: a pick per look. - Cast the characters (
nano-banana-character-sheet, four designs each, the picked frame as the style reference), then redraw the picked style frames with the cast and make text-free copies of them for every later shot. Gate: a pick per character. - Shot list on the bars (
guide-time-a-video-to-a-song, Process step 5), with the look, the baked lettering and the overlay idea per shot. Gate: the shot list. - Stills, one nano banana Pro edit call per shot at 2K, the prompt in
references/shot-still-prompt.md(guide-pick-a-stylized-look, Process step 4). Sheet every still (python-contact-sheet,images) and re-roll the misses before the person sees them. Gate: every still. - Animate each still (
guide-animate-a-shot), at least 24 frames at 24 fps so the cut can trim. Every brief keeps the frame's drawn style, mouths closed, lettering as drawn and no new text, one continuous shot, the characters on model. Gate: both models on the first minute of the song, cut on the bars, and the person picks the model (Choices); then the rest, shown the same way. - Cut on the bars (
guide-time-a-video-to-a-song, Process step 7): one look's shots can be stepped on twos. Gate: the cut with the song, before captions. - Kinetic lyrics (
hn4a-compositor-kinetic-lyrics), skipping the lines baked into a shot. Gate: the live spec (hn4a-compositor, Live spec, then render), with stills of every special moment; fix a spot after the render with--into. - Deliver (
guide-prepare-renders-for-delivery): the 1080p H.264 at CRF 19, the render master, and a 720p small copy (halftone compresses poorly: 332 MB at 1080p for 3:22).
Choices this workflow makes
- The bar-grid route: the shots, the cut and the captions all read the same downbeats.
- Two caption layers planned in the treatment: lyrics drawn into the world where the look can carry them, kinetic type for the rest.
- H3 Max for this cartoon video. On cartoon shots H3 Max and Kling v3 Pro split about evenly: H3 was better at big physical action (it commits at once), Kling at keeping baked lettering and graphic detail intact (H3 warped a carved line once; Kling broke a word once). The person picked H3 for every shot: cheap, fast, and its motion read best cut together. Run the same one-minute test on a new look.
Dead ends
- Animating before casting: designs drifted between shots until every character had a model sheet.
- Baked text written as a note: "TITLE + THE END stamp" was printed literally. Write the exact lettering.
- Kling for big cartoon action: it idled, then snapped, and drifted into odd close-ups at the end.
- Characters off model in about 1 in 12 stills (other animals, a human, outfits
added) until a rule line on real anatomy and "only the model-sheet outfit" went into
every prompt (
references/shot-still-prompt.md).
Worked example
A 3:22 pop song at 4/4, two cartoon animal characters (Character A sings, Character B is "you"), each with their own look (an ink comic with coarse halftone and misregistered plates; a xerox-and-spray-paint collage), the two looks clashing on the bridge and merging on the last chorus. 84 shots, no lip sync, about $120 of generation in all.
| Step | Result |
|---|---|
| Song map | under $1; the song file's own timestamps 5 s off, ignored |
| Style frames | 4 per look, one pick each, plus a clash look |
| Cast | 4 sheets per character; picked sheets cleaned of stray lettering |
| Stills | about $15 with re-rolls ($0.15 each at 2K, ~50 s each, 4 at once, no empty results); the anatomy and wardrobe rule lines added after the first batch |
| Animation | H3 Max about $15 for every shot ( |
| Cut | 83/83 cuts within a frame of their bars; the converted WAV ran 0.19 s shorter than the source reported, so the cut uses the WAV |
| Lyrics | good word times from the vocal stem; misheard words kept their times under the sheet's text; one chorus of three garbled, timed from another |
| Captions | 315 layers, rendered in 3:47 with 8 workers; one sound-word burst placed by eye covered a face and was moved against a still |


