AI Video Playbook

Beat driven music video

A complete walkthrough of a 40 second music video with a Suno track, a beat map, Seedance audio references and cuts on the beat.

This recipe makes a 40 second music video where every cut lands on a bar line. The music comes first: you make the track, map its bars to seconds, then generate the pictures to fit. Seedance 2.5 takes a slice of the song as an audio reference so the motion follows the rhythm. Expect about 22 USD of generation at API rates plus a paid Suno plan.

The brief

ItemChoice
Track40 seconds of an upbeat synth pop song at 120 BPM, made in Suno on a paid plan
ConceptLAN, a bicycle courier, races one glowing package across a neon city at night
CharactersOne. No singing on camera, so no lip sync.
StyleStylized anime inspired 3D, rain-slick streets, magenta and cyan neon
Length and format40 seconds, 16:9, 12 shots in 4 generations of 10 seconds
ModelSeedance 2.5 in Dreamina, with the song slice as @Audio 1

Why 120 BPM: one beat is 0.5 seconds and one bar of four beats is exactly 2 seconds. Every bar line falls on a whole second, which matches the whole second timestamps that Seedance 2.5 accepts.

Budget

ItemAmountCost
Suno Pro1 month8 USD/month on yearly billing; 2,500 credits and 20 downloads a month
Character and location sheets2 to 4 imagesA few image credits
Draft takes at 480p4 generations x 3 takes x 10 s = 120 sAbout 12.40 USD (0.103 USD/second)
Final takes at 720p4 x 10 s = 40 sAbout 9.20 USD (0.231 USD/second)
Total videoAbout 21.60 USD at Seedance 2.5 API rates

Prices checked on 15 Oct 2026. The video figures use the official ModelArk rates as a yardstick. Dreamina shows its own credit cost before you generate, and the pricing page does not say whether an audio reference changes the cost, so check the quote on your first take. Suno's monthly billing price was not in the research; check the pricing page.

Make the song on a paid plan

Songs made on Suno's free plan cannot be used commercially, and they do not become usable if you upgrade later. Songs made while you subscribe keep their rights after you cancel. The free plan also has no downloads.

Walkthrough

Idea and track

Generate the song in Suno Pro first. Ask for an upbeat synth pop track at 120 BPM in the style description, then pick the best 40 seconds (for example the end of a verse into the chorus). Check the tempo by placing beat markers in your editor before you plan any shots. If the tempo is off, every cut in the plan will drift.

Keep the visual idea to one character and one kind of location. Stylized, short cuts forgive small drift, which is why music videos suit AI well.

Beat map and shot list

Write the song structure as bars, then give each section a cutting rhythm: longer holds in quiet parts, quick cuts in loud parts.

Song timeSectionBarsGenerationCut every
0 to 10 sVerse end5A2 bars (4 s)
10 to 20 sBuild5B1 to 2 bars
20 to 30 sChorus5C1 bar (2 s)
30 to 40 sOutro5D2 to 3 bars
#Song timeGenFraming and cameraAction
10 to 4 sAWide, slow aerial descentRain-slick neon street at night, empty
24 to 8 sAMedium, tracking left to rightLAN rides her bicycle through the street
38 to 10 sAClose-up, staticThe package in her basket pulses with light on the downbeat
410 to 14 sBLow angle, trackingLAN stands on the pedals and speeds up
514 to 18 sBSide tracking, 85mmNeon signs streak past behind her
618 to 20 sBExtreme close-up, staticLAN's eyes, reflected lights sliding across them
720 to 22 sCWide, staticThe bike jumps a puddle, splash on the beat
822 to 24 sCOverhead, staticThe bike crosses a lit crossroads
924 to 26 sCClose-up, staticSpinning spokes leave light trails
1026 to 30 sCSlow orbitLAN rides as the street lights pulse
1130 to 36 sDWide, staticShe stops at a lit doorway and sets the package on the step
1236 to 40 sDSlow pull outShe rides away, the street goes quiet

Reference images and audio slices

Make a character sheet for LAN (four views on neutral grey, no text) and a location sheet with three views of the same street. Use the same style line in every prompt: "Stylized anime inspired 3D, rain-slick streets, magenta and cyan neon, soft bloom."

Then cut the 40 second section into four 10 second audio files, one per generation. Seedance recommends reference clips of 5 to 10 seconds. Any editor can export these; with ffmpeg:

Terminal
ffmpeg -i song.wav -ss 0 -t 10 part-a.wav
ffmpeg -i song.wav -ss 10 -t 10 part-b.wav
ffmpeg -i song.wav -ss 20 -t 10 part-c.wav
ffmpeg -i song.wav -ss 30 -t 10 part-d.wav

Seedance 2.5 accepts an audio file as a reference on its own and up to 30 seconds of reference audio in total. Seedance 2.0 needs an image or video alongside any audio.

Video prompts, one per generation

Upload LAN's sheet first, the location sheet second, and the matching audio slice. Give every reference one job: the audio is for rhythm and cut timing, not for sound to copy. Timestamps sit on bar lines.

Prompt
@Image 1 is LAN: use her face, hair, red jacket and bicycle only. @Image 2
sets the street and lighting only. @Audio 1 is for rhythm and cut timing
only. Stylized anime inspired 3D, rain-slick streets, magenta and cyan neon,
soft bloom.
[0s to 4s] Wide shot, slow aerial descent over an empty neon street at night.
[4s to 8s] Cut to medium shot, tracking left to right: LAN rides her bicycle
along the wet street.
[8s to 10s] Cut to close-up, static: the package in her front basket pulses
with light on the beat.
No subtitles.
Prompt
@Image 1 is LAN: use her face, hair, red jacket and bicycle only. @Image 2
sets the street and lighting only. @Audio 1 is for rhythm and cut timing
only. Stylized anime inspired 3D, rain-slick streets, magenta and cyan neon,
soft bloom.
[0s to 4s] Low angle tracking shot: LAN stands on the pedals and speeds up.
[4s to 8s] Cut to side tracking shot, 85mm: neon signs streak past behind her.
[8s to 10s] Cut to extreme close-up, static: LAN's eyes, neon reflections
sliding across them.
No subtitles.
Prompt
@Image 1 is LAN: use her face, hair, red jacket and bicycle only. @Image 2
sets the street and lighting only. @Audio 1 is for rhythm and cut timing
only. Stylized anime inspired 3D, rain-slick streets, magenta and cyan neon,
soft bloom. Fast cuts on every bar.
[0s to 2s] Wide, static: the bicycle jumps a puddle, water splashes up.
[2s to 4s] Cut to overhead, static: the bicycle crosses a lit crossroads.
[4s to 6s] Cut to close-up, static: spinning spokes leave light trails.
[6s to 10s] Cut to slow orbit around LAN riding as the street lights pulse.
No subtitles.
Prompt
@Image 1 is LAN: use her face, hair, red jacket and bicycle only. @Image 2
sets the street and lighting only. @Audio 1 is for rhythm and cut timing
only. Stylized anime inspired 3D, rain-slick streets, magenta and cyan neon,
soft bloom.
[0s to 6s] Wide shot, static: LAN rolls to a stop at a lit doorway, steps
off and sets a glowing package on the step, then stands still.
[6s to 10s] Cut to slow pull out: she rides away down the street and the
street goes quiet and still.
No subtitles.

Use Draft mode at 480p for 3 takes per generation, keep the best, then render the final at 720p.

Editing on the beat

Put the master song on the timeline first and lock it. Add a marker on every bar line (every 2 seconds at 120 BPM). In DaVinci Resolve, AI Detect Music Beats places beat markers you can snap to; reviewers disagree on whether it is in the free edition, so check yours. In Premiere, drop markers and use Automate to Sequence. Then place each generation and trim every cut so it lands exactly on a marker. Generated cuts are often a few frames off the bar line; fix them here, not by regenerating.

If the song section is slightly too long or short, Resolve's AI Music Editor can stretch or shorten a track to length. It works best on beat driven music.

Audio

Mute every generated clip. The only sound in the final video is the master track from Suno. Generated clips may contain their own version of the music or invented sounds, and mixing them under the master makes the beat feel loose.

Finishing

Export at 1080p for web and social. On YouTube, disclose the video as altered or synthetic content: YouTube requires disclosure when AI music is the main focus of a video. AI music can still trigger YouTube Content ID claims even when you hold the rights.

What usually goes wrong

ProblemWhyFix
Motion ignores the beatUsers report Seedance ignores the audio reference or invents sound in roughly 15 to 20% of runsMake 3 takes per generation and pick the one in time; fix cut points in the edit
Cuts land a few frames off the barTiming in generated clips is approximateTrim to the bar markers in the edit
LAN looks different in each generationFour separate generations, no shared memorySame character sheet, same style line and same upload order in all four prompts
Music video cannot be monetized or gets claimedFree plan song, or Content ID matchMake the song while subscribed to a paid plan, and expect that claims can still happen
Lip sync looks wrongSinging on camera is hardKeep the character not singing, as in this brief, or treat lip sync as a separate step

There is no independent performance data for AI music videos yet. The case for them rests on workflow advice: stylized looks and short cuts hide drift, and Seedance's audio references suit beat driven montage.

Sources