КАДР

How to make a music video with AI: from storyboard to editing on the beat

Alen founder of KADR Updated 2026-10-14 5 min read

An AI music video is assembled from 10 to 30 short scenes edited to the beat of the track. The AI does not «listen» to the song and does not make the whole video: you write a storyboard following the structure of the track (intro, verse, chorus), generate each scene with a video model and cut them together to the rhythm. The artist can be put in the frame from their photo using references. A 2-minute video of 20 scenes of 6 seconds each costs 1700 to 3400 rubles in generations depending on the model (KADR data, October 2026). Below, step by step: concept, storyboard, prompts, editing and rights.

How an AI music video works

The AI generates short shots of 5 to 15 seconds (Seedance 2.5 up to 30). A music video is 10 to 30 such shots edited to the music. So the work splits into four parts:

  1. Concept: what the video is about, what the visual style is.
  2. Storyboard: which scenes go to which parts of the track.
  3. Generation of each scene.
  4. Editing to the beat.

Step 1. Concept and style

Choose one of three:

  • A story video: a story with a character.
  • An atmospheric one: mood scenes without a plot: the city at night, nature, abstraction.
  • A performance: the artist in different locations.

Decide on the style right away: «90s film», «neon night city», «soft 3D animation». Write it the same way in every prompt, and the scenes will look like one video.

Step 2. A storyboard for the track

Mark up the track by timecodes: intro, verse, chorus, bridge, ending. Each part gets its own scenes.

Part of the track Length Scenes Character
Intro 0:00 to 0:15 2 to 3 slow, wide shots
Verse 0:15 to 0:45 5 to 6 the character, details, the story
Chorus 0:45 to 1:05 4 to 5 dynamics, bright shots
Verse 2 1:05 to 1:35 5 to 6 development
Ending 1:35 to 2:00 2 to 3 slowing down, the final shot

Rule: in the chorus the scenes are shorter and more dynamic; in the intro and the ending they are longer and calmer.

Step 3. The artist in the frame

If the video needs the artist:

  1. Make reference portraits from their photo in Nano Banana Pro in the style of the video: an AI photoshoot from your photo.
  2. Use them as the first frame (image-to-video) for scenes with the artist.
  3. For close-ups of the face, take Kling v3: it holds facial expressions better.
  4. To make the artist sing on camera to your vocals, upload their photo and a piece of the track to OmniHuman 1.5 (up to 30 seconds at a time). To make them repeat a dance or movements from your video, use Kling Motion Control.

How to keep the same appearance in every scene: character consistency.

Step 4. Scene prompts

A shared style line plus a specific scene. An example for a «neon night city» video, 9:16 format:

  • Intro: «Neon night city, rain, a top-down view of a wet street with reflections of signs, a slow camera fly-through forward, synthwave aesthetic».
  • Verse: «A guy in a black jacket walks down a wet street under neon signs, the camera moves in front of him, rain, synthwave aesthetic».
  • Chorus: «Close-up: the guy raises his head to the rain, drops fly into the camera, neon flashes, a fast camera orbit, synthwave aesthetic».

Sound from the model is usually not needed for a music video: your track goes on top. Turn sound off, on many models it is noticeably cheaper. How to write prompts: how to write prompts for video, for long scenes: prompts for Seedance 2.5.

Step 5. Editing to the beat

  • Place markers on the strong beats of the track.
  • Cut the scenes so that the cuts land on the markers.
  • Cut more often in the chorus, less often in the intro.
  • One 6-second scene often gives 2 or 3 pieces of 2 seconds: use them in different places.

Which model to choose

Scenes Model
Story, style, one character throughout the video Seedance 2.0
A long continuous shot up to 30 s Seedance 2.5
Close-ups of the artist, facial expressions, motion Kling v3
The artist sings to your vocals from a photo OmniHuman 1.5
The artist repeats a dance from your video Kling Motion Control
Atmosphere, light, nature, the city Veo 3.1 Fast or Lite

How much it costs

A 2-minute video, 20 scenes of 6 seconds, 1080p:

Model Scene 20 scenes With attempts ×2
Veo 3.1 Lite ~84 rub. ~1680 rub. ~3360 rub.
Veo 3.1 Fast ~126 rub. ~2520 rub. ~5040 rub.
Kling v3 ~168 rub. ~3360 rub. ~6720 rub.

KADR prices with sound, October 2026; without sound many models are cheaper. This is not a public offer. The most cost-effective approach is to combine: atmospheric scenes on Veo 3.1 Lite, close-ups of the artist on Kling v3.

Rights

  • Music: your own track is fine. Someone else’s only with the rights holder’s permission.
  • Faces: the artist and other people in the video only with their consent.
  • The style of someone else’s video can be repeated, specific characters and logos cannot: someone else’s brand, logo or character.

In short

An AI music video is a storyboard following the structure of the track, 10 to 30 short scenes in one style and editing to the beat. Put the artist in the frame from a photo with Nano Banana Pro and Kling v3, make the atmospheric scenes on Veo 3.1 or Seedance. A 2-minute video costs from 1700 rubles in generations.

Frequently asked questions

Can you make a music video with AI?
Yes. You write a storyboard following the structure of the track, generate each scene with a video model (Seedance, Kling, Veo) and edit it to the beat in a video editor. The AI does not make the whole video from the song; it makes individual shots.
How much does an AI music video cost?
A 2-minute video of 20 scenes of 6 seconds each on KADR: about 1680 rubles on Veo 3.1 Lite, about 2520 rubles on Veo 3.1 Fast, about 3360 rubles on Kling v3, not counting attempts (October 2026). With attempts, budget twice as much.
Can I put myself in an AI music video?
Yes. Make reference portraits from your photo in Nano Banana Pro and use them as the first frame or a reference for each scene. For close-up scenes of the face, Kling v3 works better.
Can AI lip-sync to a song?
Video models generate their own speech and sound rather than adapting to your finished track. Accurate lip-sync to vocals needs a separate model: on KADR it is OmniHuman 1.5, which makes the person in a photo sing to your audio for up to 30 seconds. In the other AI scenes of a music video people usually do without singing on camera.
Which AI should I choose for a music video?
Seedance 2.0 and 2.5 for story and style with one character, Kling v3 for close-ups of the artist and motion, Veo 3.1 for atmospheric scenes with light. You can combine models in one video.
Can I use someone else's song in an AI music video?
Only with the rights holder's permission. Your own track is fine. On platforms, someone else's music without a licence will lead to blocking or claims.

Music video scenes in one window

On KADR: Seedance 2.5 and 2.0, Kling v3, Veo 3.1 for scenes and Nano Banana Pro for shots with the artist. Vertical and horizontal formats, and the price is shown before you start. Pay in rubles, tokens never expire.

Make a music video
Alen
Founder of KADR. I listen to a lot of Russian rap and see independent artists making music videos with AI more and more often.