AI music video pipeline
How Nightshift Studio makes a full music video in about three days.
Day 1 — song and plan
- Write the lyrics around one story. Hold the title back until the final chorus. Quote only things a person has really said in public; never invent quotes.
- Generate the song (we use Suno). Pick the take with the clearest vocal and a clean ending. Commercial use needs a paid plan.
- Split the song into sections with timestamps: intro, verses, pre-chorus, chorus, bridge, outro.
- Write a shot list: one shot per lyric line or per 4–8 beats, 25–35 shots for a 3-minute song. Each shot gets: name, duration (5–10 s), scene, camera move, and the character's action.
- Lock the character first (see the
character-lockskill) before generating any story shot.
Day 2 — generate
- Use one video model for the whole film. Switching models mid-film makes the lead look like a different person.
- Paste the same identity block into every prompt.
- Run shots in parallel where your plan allows. A 6-second 720p shot takes 10–20 minutes.
- Check every shot as it lands (see
shot-qc). Re-generate failures immediately; do not "fix it in the edit".
Day 3 — edit and check
- Detect tempo and beats (see
beat-cut). Snap every cut to a beat; put big cuts on downbeats. - Vary speed gently (0.7×–1.25×) to land action peaks on beats.
- Captions: sparse, large, never over a face. Title card at the reveal.
- Export, then watch the finished file start to end at least twice before sending.
What we learned the hard way
- Scenes must change with the lyrics; one location for a whole song feels like a demo.
- A character who only acts (no singing) removes the hardest problem, lip-sync.
- Hidden failures live in the last second of a shot: check the ends.
Want it done for you? Nightshift Studio delivers a first cut within 3 days — nightshift.fans/contact.html