AI music video for your song: models, prompts and cost

A three-minute AI music video costs from $2.38 in generation through API Stock: $0.075 for the song on Suno plus $2.30 for 23 eight-second 1080p clips on Veo 3.1 Lite. A video that follows one character is about $6 at 720p on Seedance 2.0 Mini, which also takes a slice of your track as a reference for singing shots. No model returns a finished video in one go — you render short shots, usually two or three takes of each, and cut them to the song yourself.

What can AI make for a music video today?

Three different things, at three very different prices. Decide which one you need before you spend anything.

• A visualizer — $0.002. If the song was made with Suno through API Stock, one more request turns it into an MP4: the cover art inside a ring over a blurred colour backdrop, with your artist name, vertical, wide or square. It is not a filmed video, and it is all a streaming upload or a quick share needs. Song plus visualizer: $0.077.
• A mood video — from $2.30 for three minutes. City lights, weather, roads, abstract shapes: shots with no recurring person, each written as a text prompt. This is the cheapest real video, because the cheapest good model for it needs nothing but words.
• A story or performance video — about $6 and up. One character carried through the whole song, with close-ups where they sing. This needs a reference picture of the character in every shot and a model that accepts it.

What nothing in the catalog does: take an MP3 and return a finished three-minute video, find the beat for you, or put your own face on screen reliably. The longest single clip any model makes is 30 seconds, so a music video is always many short clips and one edit. That is how music videos are made with cameras, too.

How do you make an AI music video, step by step?

1. Get the song and its map. A song on Suno costs $0.075 per request and usually comes back as two versions to choose from. For another $0.002 each you get the lyrics with timings and the tempo in BPM — the two things an editor needs. Already have a finished track? Skip this step; the video models do not care where the song came from.
2. Write a shot list. Three minutes is 180 seconds: 36 five-second shots, or 23 eight-second ones. Mark each line of the lyrics as one of three kinds — scenery, story, singing — and note the second it starts on. At 120 BPM in 4/4 time one bar lasts two seconds, so an eight-second clip is exactly four bars; cut on the bar and the video feels in time even though no model heard the beat.
3. Fix the look with still pictures first. Make one picture of your character and one of each location on an image model: drafts on Nano Banana 2 Lite at $0.005, finals on Nano Banana 2 at $0.029 for 2K. A new still costs three cents; a new five-second video shot costs 17 at 720p. The image-to-video guide calls this the single biggest cost lever.
4. Render the shots. One idea per clip. Draft at 480p until the prompt is right, then render the keepers at 720p or 1080p. Always pass the same original character picture, never a frame from the previous clip.
5. Cut it together. Any editor will do — CapCut and DaVinci Resolve are free. Put the full song on the audio track, mute the sound the clips came with, and trim each shot to its bar.

Which AI model should you use for which shot?

No single model is right for a whole video. Prices are from the live API Stock catalog as of 2 October 2026:

In plain terms:

• Start with Veo 3.1 Lite for everything without a person. Ten cents buys eight seconds at 1080p, and the price is the same at 720p, so always take 1080p. It reads text only — it cannot see your character.
• Use Seedance 2.0 Mini for the character. It accepts your character picture as a reference, and at $0.033 per second a five-second 720p shot costs 17 cents — cheap enough to render three takes.
• Move to MiniMax H3 when the video will be watched on a big screen. It is the cheapest route to 1080p with a reference picture: $0.05 per second, clips of up to 10 seconds. The first five pictures per request are free; each one after that adds $0.02.
• Pay for Seedance 2.5 only for a showpiece. At $0.215 per second at 720p it costs six and a half times what Mini does; its case is one unbroken take of up to 30 seconds. What else it adds over the 2.0 models is in our Seedance 2.5 breakdown.

How do you get a singer whose lips follow the song?

A text prompt knows nothing about your melody. The only route to a mouth that moves with your vocal is a model that accepts audio as a reference. Several in the catalog do; these four cover every budget:

You give the model three things: the picture of your singer, a short slice of the song, and a prompt that says who is singing. Seedance 2.0's developers describe it as built for "four input modalities: text, image, audio, and video" in their technical report, with up to three audio clips per request — the same limit you see above. To cut the slice from a Suno song, the crop action keeps any range of seconds for $0.012; for your own track, export the slice from any audio editor and put it somewhere with a public link.

Keep your expectations honest. Our text-to-video guide says of spoken lines that lip sync "is decent at conversational pace and falls apart on long sentences", and singing is harder than speech. So build the video the way editors always have, with the singing close-up as an accent rather than the backbone:

• One line of the song per shot, five to eight seconds, then cut away to story or scenery.
• Close-up, facing the camera, camera locked. A wide shot leaves the mouth too few pixels to matter.
• Test at 480p first. Five seconds on Seedance 2.0 Mini costs $0.08. Render at 720p only the takes that hold.
• If the mouth follows the drums instead of the words, try a reference with the vocal alone: Suno's stems action separates the lead vocal for $0.15.
• Use shots that cannot go out of sync. The singer from behind, hands on the instrument, a silhouette against a window. Half of the performance shots in a filmed video are these anyway.

For a dance, there is a different tool. Film yourself or a friend on a phone for 3 to 30 seconds, and Kling 3.0 Motion Control makes your character picture perform that movement, at $0.065 per second at 720p. It is built to transfer body movement and gesture; check the face before you count on it.

Which prompts work for a music video?

One running example: a dreamy synth-pop song about the last night bus home, with an animated girl in a yellow raincoat. The trick that holds a whole video together is a style sentence you repeat word for word in every prompt — here, "hand-painted 2D animation, muted teal night, warm orange lights".

The song — Suno, simple mode:

Dreamy synth-pop at 120 BPM about riding the last night bus home through a
rainy city, soft female vocal, warm analog pads, a chorus that lifts

You get complete songs with lyrics written for you, usually two versions. Suno decides the length; to steer it, switch to custom mode with your own lyrics and a target duration of 10 to 360 seconds. The music generation guide covers styles and lyric sheets.

The character picture — Nano Banana 2, 16:9, 2K:

Character design for an animated film: a girl with short dark hair in a
bright yellow raincoat, standing at a bus stop at night in the rain.
Hand-painted 2D animation, muted teal night, warm orange lights.

This picture goes into every character shot. A drawn or animated look is the safer choice: reference pictures of realistic people are often rejected by the models' safety checks (the charge is refunded, but you get no clip).

Scenery — Veo 3.1 Lite, 16:9, 1080p, $0.10:

A night city seen through the rain-streaked window of a moving bus, shop
signs sliding past and smearing into long streaks of colour, slow steady
drift from left to right, focus on the raindrops. Hand-painted 2D
animation, muted teal night, warm orange lights.

Name the camera move, as here, or the model invents one. Describe what you want rather than what you don't: "no text" in a prompt can summon text just as easily.

Story shot — Seedance 2.0 Mini, character picture attached, 5 seconds, 720p, $0.17:

The girl from the reference image sits by the window of an empty night
bus and turns her head toward the glass. Slow push-in. Street lights
sweep across her yellow raincoat. Hand-painted 2D animation, muted teal
night, warm orange lights.

One action and one camera move. Two actions in five seconds is where a character starts to change shape mid-shot; make it two clips and cut between them.

Singing close-up — Seedance 2.0 Mini, character picture and a 6-second slice of the chorus attached, 720p, $0.20:

Close-up of the girl from the reference image by the bus window. She
sings the line from the reference audio to the camera, small natural
head movement, camera locked off, street lights passing behind her.
Hand-painted 2D animation, muted teal night, warm orange lights.

Faces and hands are the first things to drift, so keep these shots short and place each one in the edit at the exact second its slice was cut from.

How much does an AI music video cost?

Footage is nearly the whole bill. For 180 seconds, with prices as of 2 October 2026:

The second column is the honest one. You will not keep every render, and three takes per shot is a reasonable plan for a first video. A mixed plan is cheaper than any single premium model. For the night-bus example:

With three takes of every video shot, the same plan comes to about $12.28. Generations that fail are refunded automatically; clips that finish but you do not like are paid for, which is why the 480p draft matters.

For comparison, the dedicated music-video apps that We Rave You tested in August 2026 are subscriptions, most of them listed at $26 to $39 a month. They analyse the beat, build a storyboard and assemble the cut for you. If you want to upload a song and get a finished video with no editing, one of those is the better tool. Paying per shot wins when you want to direct: choose the model for each shot, keep only what you like, and pay nothing in a month you release nothing. API Stock is prepaid with one balance for every model — see pricing, or line up the video models on the compare page and in our video pricing comparison.

What are the limits and trade-offs?

• You are the editor. The models return clips; the cut, the timing and the titles are yours. Budget an evening for a first video, most of it choosing takes.
• The character can drift. Hair length, the shade of a coat, the shape of a face. One fixed reference picture and one repeated style sentence reduce it; short shots hide it.
• Real faces are a problem. If you want yourself in the video, expect rejections on photo references. A drawn character, or motion transferred from your phone footage, is the workable route today.
• 480p is for drafts. It is fine for checking a prompt and too soft for a release. 720p holds up on a phone; take 1080p for anything watched on a television.
• Clips arrive with their own sound. You will mute it under the song. On Kling 3.0 switching sound off also lowers the price, from $0.135 to $0.09 per second at 720p; on Seedance the price is the same either way.
• Download what you keep. Result links are for delivery, not storage; save every approved clip the day you render it.
• Check the rules before a commercial release. AI music is moving fast: Suno replaced its models with v6 on 9 September 2026, as TechCrunch reported, while lawsuits over training data continue. Read your distributor's and your video platform's rules on AI content, and only upload audio you have the right to use.

Can you automate it?

You do not need code for a first video: every model page has a playground where you paste the prompt and the links to your picture and audio. When you want to render a whole shot list from a spreadsheet, every shot is one request of the same shape. This is the singing close-up from above — 6 seconds at 720p, $0.20:

curl -X POST https://api.api-stock.com/api/v1/generation/create \
  -H "Authorization: Bearer $API_STOCK_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.0-mini",
    "input": {
      "prompt": "Close-up of the girl from the reference image by the bus window. She sings the line from the reference audio to the camera, small natural head movement, camera locked off. Hand-painted 2D animation, muted teal night, warm orange lights.",
      "imageUrls": ["https://example.com/raincoat-girl.png"],
      "audioUrls": ["https://example.com/chorus-slice.mp3"],
      "duration": 6,
      "resolution": "720p",
      "size": "16:9"
    }
  }'

It returns a task id at once; the finished MP4 arrives by polling or webhook. The fields for each model are in the video model docs, and all 31 Suno actions, including crop, stems and the music-video visualizer, are mapped in our Suno API guide.

FAQ

How much does an AI music video cost?

Through API Stock, as of 2 October 2026, three minutes of footage costs from $2.30 on Veo 3.1 Lite at 1080p to $38.70 on Seedance 2.5 at 720p, one take per shot; the song adds $0.075. A mixed plan with one recurring character is about $4.33, or about $12 with three takes of every shot.

Can AI make a full music video from a song in one go?

Not with the models in the catalog: the longest single clip is 30 seconds, and none of them takes a whole song as input. You render short shots and cut them to the track in an editor. The one-request exception is Suno's visualizer, an MP4 with the cover art for $0.002.

Which AI model is best for music videos?

Veo 3.1 Lite for scenery and mood shots ($0.10 per 8-second 1080p clip), Seedance 2.0 Mini for shots with a recurring character ($0.033 per second at 720p), and MiniMax H3 when you need that character in 1080p ($0.05 per second).

Can AI lip-sync a singer to my song?

Partly. Seedance 2.0 Mini, MiniMax H3, Wan 3.0 Video and Seedance 2.5 accept a short slice of your track as an audio reference next to the character picture. Expect it to work for one line in a close-up, not for a whole verse, and test at 480p before paying for the final.

Can I use my own song instead of one made with Suno?

Yes. The video models only need prompts and pictures, and the audio reference is any short file with a public link. The Suno extras — the $0.002 visualizer, timed lyrics and tempo — start from a song generated with Suno on your account; for an outside track, find the tempo and mark the lines in your editor instead.

Do I need to know how to code?

No. Each model page on API Stock has a playground for prompts, pictures and audio links. You do need a video editor to join the clips; code only matters when you want to render a long shot list automatically.