A lyric video on 8frame is two jobs in two places: AI backgrounds made on the board (one look per song section, as stills turned into short clips), and the lyrics added in the Video Editor as caption layers from text you paste. The limit people do not expect is timing. The Video Editor's "Generate captions" spaces your pasted lines at an estimated 160 words per minute; it does not listen to the song, so every line still has to be moved by hand to where it is sung.
This page is the production recipe. If you are choosing between ready-made lyric video apps, the comparison is in lyric video makers compared. If you want a narrative music video rather than words on moving backgrounds, start with how to make a music video with AI.
TL;DR
| Step | Where on 8frame | Price in credits | Input |
|---|---|---|---|
| Song | Upload, or a music node | Upload: none. ElevenLabs Music 81 per started minute; MiniMax Music v2 5 per generation (needs your lyrics) | MP3 or WAV, or a prompt plus lyrics |
| Look per section | Image node: Nano Banana 2 / Nano Banana Pro / Seedream 5 | 11 at 1K / 21 at 1K or 2K / 6 per image | A style prompt, optional cover art as reference |
| Loop clip per section | Kling 2.5 Turbo Pro (start and end frame) | 48 for 5 s, 95 for 10 s | The section still, used as first and last frame |
| Long, calm section | Wan 3.0 / Seedance 2.5 | 9.45 per second at 720p, 2 to 30 s / 157 for 5 s at 720p, up to 30 s | One start image |
| Lyrics | Video Editor caption layer | not a generation | Lyrics pasted as text |
| Song under the picture | Video Editor sound layer | not a generation | Your audio file |
A credit is about $0.01 at pack rate. The Video Editor work (captions, text, sound, cutting) is editing, not generation.
Why the lyrics never go into the image
It is tempting to prompt an image model for "the words of the chorus in neon". Do not. A generated image bakes the text in: if a word comes out misspelled you pay for a new image, you cannot change the timing, the font or the size, and a translated version needs every frame remade. On the Video Editor timeline a lyric is a text layer you can edit, move and restyle for free, over a background that never changes.
So the board makes pictures without words, and the timeline adds the words. Write "no text, no letters, no signage" into every background prompt; there is no negative-prompt field on any 8frame node, so the exclusion has to be in the prompt itself.
The board, node by node
There is no lyric video template on /workflows yet. Lay the board out as one row per song section: intro, verse 1, chorus, verse 2, chorus, bridge, last chorus, outro. Repeated sections can share a row.
1. Put the song on the board
Upload your track (MP3 or WAV) to the board so everyone working on it hears the same file. If you need a track first, two music nodes fit a lyric video:
- MiniMax Music v2 costs 5 credits per generation and requires lyrics of 10 to 3,000 characters, so it makes sung songs from words you wrote.
- ElevenLabs Music bills 81 credits per started minute (3 seconds to 10 minutes); a 30-second track still bills a full minute.
If the song is someone else's, the lyrics belong to its writers and publishers. A lyric video displays them in full, so get permission for that use before you publish, separately from any licence for the recording.
2. Write the section notes
On a sticky note per row, write the section's lines, its length in seconds (from the track) and one mood word. The lengths decide which video model can cover a section in one clip, and the mood word keeps the look of each section different enough that the viewer feels the song change.
3. One look per section, as a still
Generate a background still per section with one shared style line ("grainy 16 mm film, dusk colours, shallow depth of field") plus the section's own subject. Attach your cover art as a reference so the video belongs to the release. Nano Banana 2 is 11 credits at 1K and 17 at 2K; Nano Banana Pro is 21 at 1K or 2K; Seedream 5 is 6 and takes up to 14 reference images.
Leave space. Lyrics sit in the lower third or the centre, so prompt for a quiet area there ("empty sky in the lower half", "soft out-of-focus floor").
4. Turn each still into a clip that can repeat
A verse can run 30 seconds or more, and most video models stop well before that. The cheapest way to cover a long section is a loop: a short clip that ends where it began, repeated on the timeline.
On the canvas, the models that accept an end frame are Kling 2.5 Turbo Pro, Kling 2.1 Pro, Kling O1 Image and Kling O3, plus Seedance with a start image. Put the same still in the start and end slots and ask for slow, returning motion: drifting smoke, a slow camera push that eases back, light moving across a wall. Kling 2.5 Turbo Pro is 48 credits for 5 seconds and 95 for 10. Kling v3 does not take an end frame on the canvas, so it cannot close a loop this way.
For a section that should not loop (a bridge that builds, an outro that fades), use one long take instead: Wan 3.0 runs 2 to 30 seconds at 9.45 credits per second at 720p; Seedance 2.5 runs up to 30 seconds (157 for 5 seconds at 720p, 937 for 30 seconds).
5. Build the timeline
Open the Video Editor mode. Put the song on a sound layer first, then place each section's clip under its part of the song, repeating loops as needed. Then add the lyrics.
Timing the lines by hand
This is where most of the hours go, on any tool that does not transcribe.
- Paste the lyrics of one section into "Generate captions". You get a caption layer with the lines spaced at an estimated 160 words per minute.
- Play the song and drag each line so it appears just before it is sung. Sung lines are rarely at speaking pace: a held note can take four seconds for one word, a rap verse can run far faster than 160 words per minute.
- Repeat per section. Pasting section by section keeps the drift small, because each block starts at the right place.
- Keep one line on screen at a time for most songs, two for fast verses. Leave each line up until the next one starts, so the screen is never empty mid-phrase.
The Video Editor has no audio transcription and no SRT export. If you need a subtitle file for a platform, write it in a subtitle tool from the same lyric text.
For the look of the text, the Video Editor's text layers handle font and placement; for the backgrounds, its Filter Preset menu (Retro, Vintage, Film Noir, Super 8, Cinematic and others) and a per-clip Brightness slider can pull clips from different models into one grade. There are no curves, colour wheels or LUTs.
Which model for which section
| Section need | Model on 8frame | Lengths | Price in credits | Note |
|---|---|---|---|---|
| Clean loop | Kling 2.5 Turbo Pro | 5 or 10 s | 48 / 95 | Start and end frame on the canvas |
| Loop with sound off, more motion control | Kling O3 | 3 to 15 s | 57 silent for 5 s | Needs a start image; end frame on the canvas |
| One long calm take | Wan 3.0 | 2 to 30 s | 9.45 per second at 720p | Start image or text only |
| One long take, 720p ceiling | Seedance 2.5 | 4 to 30 s | 157 for 5 s at 720p | Tops out at 720p |
| Cheap draft of a section | Pruna p-video Draft | 1 to 20 s | 4 for 5 s at 720p | We quote price and settings, not results |
| 2K background | Hailuo H3 | 5, 10, 15 s | 88 / 176 / 264 | Files come with an audio track: mute it under your song |
Many video models now make sound by default (Seedance 2.5, Hailuo H3, Wan 3.0 Prime, Kling v3 with its toggle on). Under a song you want none of it: mute those clip layers on the timeline, or pick silent settings where the node offers them.
Where it breaks
- Timing is manual. Caption spacing is an estimate from text, not from the vocal.
- Montage drops sound. If you join section clips with the Montage node first (2 to 10 clips, at most 15 s of each, 10 credits), the merged file has no audio. That is fine for backgrounds, but the song itself goes on the Video Editor sound layer.
- No video upscaler. A 720p Seedance clip stays 720p. Pick the output resolution at generation time.
- No 4K video. No video model on 8frame outputs 4K.
- Baked text. If an image model writes letters into a background anyway, regenerate or crop; do not try to cover it with your lyric.
- AI disclosure at release. If you put the video out through a distributor, AI-generated visuals are their own question on the upload form; see distributor rules on AI music and visuals.
FAQ
Can 8frame sync lyrics to my song automatically? No. "Generate captions" turns pasted text into a caption layer timed at an estimated 160 words per minute. It does not transcribe audio, so you move each line to the vocal by hand.
Can I make the whole lyric video in one generation? Not on 8frame, and you would not want to: baked-in lyrics cannot be retimed or corrected. Generate backgrounds, then add lyrics as layers.
Which aspect ratio should a lyric video use? 16:9 for a YouTube upload, 9:16 for Shorts, Reels and TikTok. Choose it on the image and video nodes, because the backgrounds are generated in that shape.
How do I cover a long verse with short clips? Make a loop: the same still as first and last frame on an end-frame model such as Kling 2.5 Turbo Pro, then repeat it on the timeline. Or use one long take on Wan 3.0 or Seedance 2.5, up to 30 seconds.
Can I use a song I do not own? Only with permission. A lyric video shows the full lyrics and plays the full recording, which are both someone else's rights unless the song is yours.
Sources
- 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.