What do you need before Kaiber can turn your track into a music video?

BEFORE OPENING KAIBER
[x] final audio file, exported once, never touched again
[x] section map with timestamps (intro, verse, chorus, bridge, outro)
[x] one visual line per section, taken from the words of that section
[x] one style anchor + one reference image, reused everywhere
[x] the lyric, in a text file, with the time each line starts
[ ] budget for reruns (every retry costs credits)

That sheet is the answer. Kaiber needs three things from you before it can do anything useful with a track: a finished audio file you will not change, a visual brief written section by section, and the lyric with its timing. Everything else, the look, the motion, the camera, gets decided inside the tool, but it gets decided against those three inputs. Skip one and you pay for it in reruns.

The reason is simple and a little unfair. Kaiber animates. It takes the file you give it and turns a description, an image or your own clips into moving pictures cut to the sound. What it does not do is decide what the song is about. If you arrive with a track and a vague idea ("something dreamy, blue, kind of cinematic"), the tool will happily spend your credits producing a minute of dreamy blue motion that has nothing to do with the second verse, where the song actually turns.

If you are still choosing between video tools rather than working in this one, read our shortlist of AI music video generators first; this page assumes you have picked Kaiber and want the first render to count. If your track came out of Suno and you want the wider picture of the three routes from audio to video (visualizer, AI clips, real footage), that lives in From Suno track to music video.

What Kaiber is, as far as its own pages say

We are a different company from Kaiber, and its plans, credit costs and interface change often enough that anything precise we wrote here would be stale by the time you read it. So here is only what is stable and documented in Kaiber's own help centre at the time of writing.

Kaiber describes itself as a suite built for musicians and the people who market them. At the time of writing its help centre splits the product into three tools: Canvas for generating video, images and audio; Beat Sync, which takes your track plus your images or clips and cuts them into videos synced to the beat; and Editor, a timeline for text, transitions and final assembly. For a whole song there is also a workflow called Music Video Montage: you give it the audio and a short description, optionally a reference image, and it builds the video from what it hears in the music. The product has been reorganised at least once (the help centre has a whole section on what changed), and generation runs on credits that come with a plan.

For file formats, maximum length, resolution, watermarks and commercial-use terms, go to Kaiber's current help and pricing pages. Those are exactly the numbers that move, and a wrong one here would cost you more than no number at all.

Where we fit, and where we do not

sunomarket makes words, not pictures and not sound. The lyrics generator takes a short form (who is speaking, which language, which structure, which emotion, what the song is about in up to 300 characters) and returns a title and a full lyric with [Verse] and [Chorus] section tags. One generation is 20 credits, which is $0.20 at 100 credits to the dollar. Credits come only with a plan and there are no free generations: the $5 monthly plan brings 500 credits a month, which is 25 lyrics, and the $19 lifetime option, paid once, brings 1 500 credits a single time, which is 75 lyrics with no refill.

We do not sing, we do not make audio, we do not render video, and we have nothing to do with Kaiber or Suno beyond the fact that people use all three in a row. The order matters, though, and it is the reason this page exists: Kaiber starts where the song ends, and the song starts with words. If you have a finished track, skip to the next section. If you have an idea for a video and no song yet, the cheapest step in the whole chain is the first one, and it is also the one that decides what every later frame is about.

Lock the track before you upload anything

Audio-reactive video is married to the exact file it was made from. Every pulse, every cut in motion, every flare of light is tied to a moment in that waveform. Change the file afterwards and the marriage breaks: trim two seconds off the intro, and every beat in the video now lands two seconds early. Swap in a remaster with a louder chorus, and the motion no longer matches the peaks.

So the rule is: finish the song, then make the video. In practice that means four checks.

It is the final version. Not "final unless I redo the bridge". If there is any chance of another pass on the vocal, do it now. A regenerated song from Suno or anywhere else is a different file with different timing, even when it sounds almost the same.

It is trimmed the way it will be released. If the release starts on the downbeat of the first verse, cut the upload to start there too. Silence at the head of a file is time the video has to fill with something.

It is the best export you have. A clean, full-quality file gives the reactive engine clear peaks to follow. A heavily compressed copy ripped from a streaming preview gives it mush.

You know its exact length. Write it down to the second. You will need it for the section map, and you will need it when a clip comes back short.

One more thing to decide before you upload: whether the video covers the whole song or only a piece of it. A 15 to 30 second segment for a vertical teaser is a different job from a full three-minute clip, in cost and in planning. For a looping clip behind a streaming release, see Spotify Canvas with AI, which is its own format with its own rules.

Map the song before you describe a single image

The section map is the most useful ten minutes of the whole project. Play the track with a timer and write down where each part begins. Here is a typical shape for a mid-tempo song of three minutes and twelve seconds:

0:00  intro       piano alone, quiet
0:14  verse 1     vocal enters, drums light
0:46  chorus 1    full band, loud
1:10  verse 2     back down, bass carries it
1:42  chorus 2    loud, adds strings
2:06  bridge      drums drop out, vocal close
2:30  chorus 3    loudest point, everything in
2:58  outro       piano again, fade to 3:12

This map does two jobs. First, it shows you where the energy changes, and audio-reactive visuals live on energy. The jump from the bridge into the last chorus at 2:30 is where the biggest visual change belongs; the drop at 2:06 is where the motion should almost stop. If you prompt one continuous look for the whole song, the reactivity will still pulse along, but nothing will change at 2:30, and the moment the song is built around will look like every other moment.

Second, the map tells you how many separate generations you are really planning. How many depends on the route. Kaiber's help centre describes Music Video Montage as one description for the whole track (one subject, one setting, one style), so there the map shapes that one description and tells you where to cut later. If you build scene by scene in Canvas, eight sections is not necessarily eight prompts (a verse and its chorus can share a scene that simply gets louder), but it is rarely one. Knowing the count before you start is how you estimate credits before you spend them.

A flat track is a warning sign here. If you cannot hear any of these boundaries, the reactive engine will not find them either, and the whole video will breathe at one pace. That is a problem with the song, not the video, and it is cheaper to fix in the song.

Write the visual brief from the lyric, not from the genre

This is the step where a video goes generic. People describe the genre ("synthwave, neon, retro city") instead of the song. Every synthwave track then gets the same neon city, and yours becomes indistinguishable from every other track fed the same prompt.

The fix is to take each visual line from the words sung in that section. Here is a verse and chorus from an example song about someone left behind in a coastal town:

[Verse 1]
The ferry leaves at six, I hear it from the bed
Your jacket on the hook, the radio half dead
I fold the map along the lines you used to trace
And leave the porch light on, in case

[Chorus]
Salt on the window, salt on my hands
Nobody here ever understands
Why I keep waiting where the water ends
Salt on the window, and the light still on

And here are the visual lines that come out of it, one per section, each using only what the words above give you:

verse 1  : ferry pulling away from a small pier, seen through a bedroom window, a jacket on a hook in the foreground
chorus 1 : a window crusted with salt, a porch light glowing behind it, waves at the end of a street

Almost nothing in those lines was invented: the ferry, the bed, the jacket, the salt, the porch light and the water's edge are all in the lyric. Only the framing is added: a pier, a window, a street, where the camera stands. That is the test. If a reader could hold your visual line next to the lyric and point to the word each image came from, the video will look like it belongs to this song. If they cannot, it will look like it belongs to the genre.

Two practical notes. Keep each visual line to one scene; a prompt that asks for a ferry, a city skyline, a thunderstorm and a woman crying gets you a blur of all four. And describe things the camera can see, not feelings. "Loneliness" is not a picture. A single lit window on a dark street is.

Pick one anchor and never let go of it

The second way AI music videos fall apart is drift. The verse is painterly watercolour, the chorus comes back photoreal, the bridge looks like a video game. Each clip is fine alone and the whole thing looks like five different videos glued together.

Decide the look once and repeat it word for word in every prompt:

style anchor: muted film photograph, cold morning light, grey-blue palette, soft grain

Then paste that exact line at the end of every section's visual line. Do not paraphrase it between prompts ("cold light" in one, "chilly dawn tones" in the next); to a model those are two different requests.

At the time of writing, Kaiber's Music Video Montage has Reference and Restyle modes that take an image, and Beat Sync is built from the images and clips you upload (the options have changed before, so check the help centre). A single reference image is an even stronger anchor than words. The obvious candidate is your cover art, because the video and the cover should look like the same release anyway. If you do not have a cover yet, album cover generators covers what makes one work at thumbnail size, which is also the size most people will first see your video at.

Keep the lyric beside you even if no words appear on screen

It is tempting to think the lyric only matters for a lyric video. It matters for every video, because it is the most exact timing document you have. The section map says where the chorus starts; the lyric with timestamps says where the line "and the light still on" lands, which is where the porch light in your chorus scene should flare.

A timed lyric looks like this, and it takes one listen with a pause button:

0:46  Salt on the window, salt on my hands
0:52  Nobody here ever understands
0:58  Why I keep waiting where the water ends
1:04  Salt on the window, and the light still on

If you do want the words on screen, plan to add them in a video editor after the Kaiber render rather than asking the generator to paint them into the picture. Image and video models are often unreliable at drawing legible text, and you will want control over when each line appears anyway. At the time of writing, Kaiber's Beat Sync also has a toggle that adds lyrics synced to the track, with every line editable; your timed lyric file is what you check those lines against. How to time lines so viewers can sing along (a line should appear slightly before it is sung, not on the first syllable) is covered in Can you make a lyric video before you have the lyrics?.

Budget the reruns before you start

Every generation costs credits, and the second, third and fourth attempts cost the same as the first. The preparation above is mostly a way of making the first attempt closer to right. A rough planning habit:

• count the separate scenes from your section map;
• assume each one will need at least one retry;
• render one short test section first (the chorus is the best test, since it is the loudest part and repeats) before committing to the whole song.

A test chorus that looks wrong tells you the anchor or the brief is off, and fixing it costs one clip rather than eight. Kaiber's help centre says the cost of a generation depends on its type, the model and the duration, and shows the number next to the generate button, so read it on your test clip before you plan the whole video.

Rights, before you publish

You need the right to use the track you upload. If the song came from a music generator, that generator's terms for your plan decide what you may do with the audio. Kaiber's own terms for your plan decide what you may do with the video it renders, including commercial use. Both change, so read the current versions rather than a forum summary.

For the words: our terms allow commercial use of the lyrics, but the rights are not exclusive. The same generator can produce similar or overlapping lines for another person, and we do not claim authorship over the output or clear it against anyone else's rights. Our disclaimer sets this out in full.

This is general information, not legal advice. If a release carries real money or a label deal, ask a lawyer in your own country.

No song yet? Start two steps earlier

A lot of people search for Kaiber with an idea for a video and no track. The order that wastes the least money is: words first, then audio, then video. Here is a brief you could put into our form for the coastal-town song above:

POV        first
language   en
structure  classic
hook       title-line
vocal      female
emotion    longing, loneliness
age        25-34
theme      A woman in a small coastal town keeps her porch light on
           for someone who left on the morning ferry. Salt on the
           windows, his jacket still on the hook. She will not leave.

What comes back is a title and a complete lyric with section tags, ready to paste into Suno or whichever tool you use to sing it. It will not be the example verse above word for word; it will be its own lyric built from that theme. Because your theme names concrete objects (a porch light, a ferry, salt, a jacket), the lyric has concrete objects in it, and those objects become your Kaiber scenes later. A theme like "a sad song about missing someone" gives you a sad, abstract lyric, and an abstract lyric gives you nothing to film.

The move

If the track is finished: export it once, map it, write one visual line per section from the lyric, fix a style anchor, and render the chorus as a test before the rest.

If the track is not finished, the video is not the first problem. Write the song around things a camera can see. Open the lyrics generator, put the objects of your video into the theme field, and generate a draft; inside a plan one lyric costs $0.20. The cheapest way in is the $5 monthly plan, which covers 25 lyrics. If you would rather pay once, the $19 lifetime option keeps the catalogue and the prompt generator open for good and comes with 75 lyrics, once, with no refill.