If the song is already mixed and mastered, the music video is a picture problem: a beat map of the track, a look, and clips cut to its bars. On one 8frame board that means a frame per song section with the bar count written on it, image nodes for the key frames of each section, video nodes whose lengths land on whole bars, and the Video Editor timeline where your real track sits on a sound layer under the clips. The limit people do not expect is that the canvas's Montage node is the wrong place to finish: it drops all sound and uses at most 15 seconds of each clip, so a music video joined there comes out silent and cut short.
This recipe starts from your own track. If you still need the song, that is a different job; how to use ElevenLabs Music and how to use MiniMax Music cover generating one. For the creative side of a music video from scratch, see how to make a music video with AI.
TL;DR
| Step | Node or model on 8frame | Fixed price in credits | Input it needs |
|---|---|---|---|
| Beat map | Frames and sticky notes on the board | 0 | The song's tempo and structure |
| Look and key frames | Nano Banana 2 or Seedream 5 Pro | Nano Banana 2: 11 at 1K, 17 at 2K; Seedream 5 Pro: 7 at 1K, 13 at 2K | Prompt plus references (up to 10 on Nano Banana; 14 on Seedream 5 Lite) |
| Clips cut to any length | Wan 3.0 | 9.45 per second at 720p, 2 to 30 s (5 s 48; 480p 24, 1080p 95 for 5 s) | Prompt or start image |
| Clips cut to any length, sound always on | Grok Imagine | 9.45 per second at 720p, 6.75 at 480p, 1 to 15 s | Prompt or image |
| Long section in one take | Seedance 2.5 | 30 s: 417 at 480p, 937 at 720p | Start image |
| Close-up performance | Pruna Avatar | 3.375 per second at 720p, up to 35 s | Portrait plus an audio excerpt |
| Final cut with the real track | Video Editor mode | Not a priced tool | Clips, your song file |
Step 1: the beat map
Write the song's structure into the board before you generate anything. One frame per section, left to right: intro, verse 1, pre-chorus, chorus, verse 2, chorus, bridge, final chorus, outro. Inside each frame, a sticky note with:
- Start time in the track (00:42).
- Bars and seconds. In 4/4, one bar lasts 240 divided by the tempo in seconds. At 120 BPM a bar is 2 seconds, so an 8-bar verse is 16 seconds; at 90 BPM a bar is about 2.67 seconds and 8 bars is about 21 seconds.
- The image idea for the section, in one line.
- Cut rhythm: a cut every 2 bars in the verse, every bar in the chorus, one long shot in the bridge.
Arrows between frames stay attached when you move them. The board has sticky notes in five colours, frames, shapes and arrows but no table, so keep the timings in the notes.
This map is what makes the video feel cut to the song. Clips that start and end on bar lines land on the beat when you lay them on the timeline; clips of arbitrary length have to be trimmed, and trims in the middle of a movement are the cuts viewers feel as wrong.
Step 2: one look for the whole song
Before any section, fix the look: a palette, a location family, a performer or no performer. Generate three or four style stills from one prompt and reference set and keep the one that fits the song. Save it as the reference for every key frame that follows. A mood board for a music video is the planning version of the same step.
If a real artist appears, use their own photos with their agreement for this use. Do not prompt for the likeness of any real musician who has not agreed.
Step 3: key frames per section
Inside each section frame, one image node per planned shot. A verse with a cut every two bars at 120 BPM over 16 seconds is four shots; a chorus with a cut every bar is eight. Use the look still as a reference on every node so the sections belong to one video.
Nano Banana 2 (11 at 1K) takes up to 10 reference images and Seedream 5 Lite (6) up to 14, the most on the roster; Seedream 5 Pro is 7 at 1K and 13 at 2K. Generate at the delivery ratio. A 16:9 video for YouTube and a 9:16 cut for Shorts or Reels are two sets of key frames, not one set cropped.
Step 4: clips that land on bars
Pick video models by the lengths they allow, because the bar math decides the length.
- Per-second models let a clip be exactly 2, 4 or 8 bars. Wan 3.0 runs 2 to 30 seconds and bills per second (9.45 per second at 720p); Grok Imagine runs 1 to 15 seconds at 9.45 per second at 720p and always generates audio, which you mute on the timeline. Kling v3 Standard and Pro render any whole second from 3 to 15, billed per second (Standard: 86 for 5 s with sound, 57 silent), and need a start image; Kling O3 also runs 3 to 15 seconds, at 11.34 per second silent.
- Fixed-length models come in set lengths: Hailuo H3 at 5, 10 and 15 s (88, 176, 264), Kling 2.6 Pro, 2.5 Turbo Pro and 2.1 Pro at 5 or 10 s. At 120 BPM, a 10-second clip is 5 bars; trim to 4 on the timeline.
- One long take for a bridge or an outro: Seedance 2.5 runs 4 to 30 seconds and bills 417 at 480p or 937 at 720p for 30 s; Wan 3.0 also runs to 30 s.
Connect each key frame to its video node. Prompt for motion that fits the section: slow drift in the verse, faster camera or subject motion in the chorus. Generated video does not know your tempo; any sync comes from where you cut.
Performance shots
A close-up of a singer mouthing the lyrics is a lip-sync job: a portrait plus an audio excerpt of the vocal. Pruna Avatar takes up to 35 seconds of audio at 3.375 per second (720p) and refuses longer audio before charging; OmniHuman v2, the OmniHuman default, is 71 per 5 s, billed per second of audio, for 4 to 35 s; Hailuo H3 Max Lip Sync clips audio at about 14.8 seconds. Cut the excerpt to the section, feed it, then lay the full song over the result on the timeline. We quote price and settings, not results: test one chorus before planning a whole performance video around them.
Step 5: the final cut with the real track
The canvas has no timeline; the Video Editor mode does, with video, text, caption and sound layers.
- Put your song file on a sound layer first. The canvas takes MP3 and WAV uploads.
- Lay clips on the video layer section by section, using the start times from the beat map.
- Trim each clip on a bar line. Mute generated audio from Grok Imagine, Hailuo H3 or Seedance; their files carry an audio track.
- Lyrics, if you want them, go on a caption or text layer. "Generate captions" times pasted text at speech pace, not your song's phrasing: place lyric lines by hand.
- Export and listen through once on headphones and once on a phone speaker.
The Montage node joins 2 to 10 clips for 10 credits a run, but the merged file has no sound and each clip is trimmed to 15 seconds at most. Use it only for a quick silent look at a section's order.
Which model for which step
| Need | Model | What decides it |
|---|---|---|
| Exact bar lengths | Wan 3.0, Grok Imagine or Kling v3 | Per second; Wan 2 to 30 s, Grok 1 to 15 s, Kling v3 3 to 15 s |
| Silent clips from a key frame | Kling O3 | 11.34 per second silent, 3 to 15 s, needs a start image |
| A 30-second section in one take | Seedance 2.5 or Wan 3.0 | Both run to 30 s; Seedance tops out at 720p |
| 2K picture | Hailuo H3 | 88 / 176 / 264 for 5 / 10 / 15 s |
| 1080p from a still | Kling 2.5 Turbo Pro | 48 for 5 s, 95 for 10 s; files carry no audio |
| Singer close-ups | Pruna Avatar or OmniHuman v2 | Up to 35 s of audio; we quote price and settings, not results |
Spotify Canvas and short cuts
A finished video can feed short loops. Spotify's Canvas guidelines ask for a clip "3-8 seconds long", "vertical 9:16 ratio", "between 720px - 1080px tall", as "an MP4 or JPG file", and advise "footage without talking, singing, or rapping" and to "avoid rapid video cuts or intense flashing graphics". So the Canvas is not a crop of your chorus montage: pick one 4- to 8-second shot without a mouth on screen, at 9:16, and keep the edges free, since Spotify notes "the edges may get cut off on some phones".
Where it breaks
- Montage is silent and trims every clip to 15 seconds at most. Finish in the Video Editor.
- Generated video has no tempo. Sync is made by cutting on bars, not by the model.
- Fixed lengths. Hailuo H3 and the older Kling models come in set lengths; trim to bars, or use a per-second model.
- Seedance stops at 720p, and there is no video upscaler on 8frame. Generate at the resolution you deliver.
- Lip-sync caps. Hailuo H3 Max Lip Sync clips at about 14.8 s; Pruna Avatar refuses audio over 35 s.
- Lyrics captions are timed at speech pace from pasted text; place them by hand.
FAQ
Can AI sync a music video to my song automatically?
Not on 8frame. The video models do not read your track. You get sync by planning clip lengths in bars and cutting on bar lines in the Video Editor, where the song sits on a sound layer.
How do I calculate clip lengths from the BPM?
In 4/4, one bar lasts 240 divided by the BPM, in seconds. At 120 BPM a bar is 2 seconds, so a 4-bar shot is 8 seconds. Per-second models such as Wan 3.0, Grok Imagine and Kling v3 can be set to that length.
Why is my merged music video silent?
The Montage node drops sound: its merged file has no audio track. Put the song on a sound layer in the Video Editor and cut there.
Can the singer lip-sync to my vocal?
Lip-sync models take a portrait plus audio: Pruna Avatar up to 35 seconds, OmniHuman v2 from 4 to 35 seconds, Hailuo H3 Max Lip Sync clipped at about 14.8 seconds. We quote price and settings, not results, so test one section first.
What are the Spotify Canvas specs?
Spotify's guidelines ask for 3 to 8 seconds, vertical 9:16, 720 to 1080 px tall, MP4 or JPG, without talking, singing or rapping, and without rapid cuts.
Sources
- Spotify for Artists, Canvas guidelines (3 to 8 seconds, 9:16, 720 to 1080 px tall, MP4 or JPG, no talking or singing footage, no rapid cuts, edges may be cut off): support.spotify.com/us/artists/article/canvas-guidelines, checked 2026-11-09
- 8frame prices, defaults and limits: the 8frame app, checked 2026-11-09
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.