Song in. Video plan out.
Give your AI video tool a feel for music. Songbrain listens to the song and tells your video model what to show, and on which beat.
- ✓Finds the strongest moment of any song
- ✓Understands what the song is about
- ✓Plans every shot on the beat, with a ready prompt
Then $0.50 per song · no subscription · works with any video model
See what Songbrain hears in a song
Four of our own songs, run through the API. Pick one: this is what your tool would get back, in plain words.
Wide establishing shot of the empty space: A weathered stone marker, a moss-covered, half-buried stone, showing age and neglect, a wide, barren field under a vast…
Slow push-in on a still world, 1.6 s.
World: A desolate, ancient battlefield at twilight, bathed in the faint glow of a rising moon, with scattered…
Every scene comes with a start image and how it moves. Short beats are moved in your edit; longer scenes are where your video model renders.
- Tempo
- 117.5 BPM
- Key
- C minor
- Genre
- Metalcore
- Virality
- 70 / 100
Feels epic, powerful, melancholic.
- #1 · 2:14–2:2969
The intense synth solo and driving beat create a massive, unforgettable moment.
- #2 · 1:43–1:5863
The driving beat and synth melody create an immediate, powerful energy.
- #3 · 0:24–0:3958
A pulsing synth line builds tension before a dramatic beat drop.
The song is about the enduring weight of past sacrifices and the responsibility to live a meaningful life in their honor.
Carried by a weathered stone marker: a moss-covered, half-buried stone, showing age and neglect → is struck and fractured by falling debris → a broken stone marker, still standing but clearly scarred.
See the full JSON →Frames rendered by us from the API's prompts with an open image model, to show what they produce. The API returns the prompts and timings; your model does the rendering.
The part of a music video your model can't do
Videos that fit the song
Your users stop getting random clips over their music. Scenes change where the music changes, and the story comes from the lyrics.
One API instead of four
Beat analysis, lyrics with timing, mood and an AI director that writes the shots come back in one call, on one bill.
Ship in days, not months
No audio team, no licence talks, no prompt tuning. Send a song and get a plan your renderer can use as it is.
Any video model
Kling, Veo, Runway, Luma, your own model: the plan is plain JSON with seconds and prompts. Switching models doesn't break anything.
Pay per song. Everything included.
Free
5 songs every month
- ✓ The full document, shot plan included
- ✓ Unlimited calls to the example endpoints
- ✓ MCP server
Pay as you go
25 credits per song · 500 credits = $10
- ✓ No subscription, no minimum
- ✓ Failed analyses are refunded automatically
- ✓ Signed webhooks, 3 songs in parallel
Volume
More than 1,000 songs a month
- ✓ Lower price per song
- ✓ Invoice, DPA, SLA
- ✓ Custom fields in the output
Limits: files up to 100 MB, songs 30 seconds to 10 minutes. MP3, WAV, FLAC, M4A, AAC, OGG or AIFF. 120 requests a minute.
Everything about your usage, in one place
Your account comes with a console at app.songbrain.ai/developers. It works with the same login and the same credits as the app.
Usage & spend
Requests, songs, credits, error rate and latency per day, compared with the period before.
Request log
Every call with time, endpoint, status, latency and key. Filter by errors, keys or MCP.
Playground
Drop a song in the browser and read the full analysis and shot plan, without writing a line of code.
Songs
Every analysed song with its status and processing time, plus the full JSON. Delete with one click.
Webhooks
Every delivery with its status and attempts, and a button to send it again.
Keys & billing
Create and revoke keys, see free songs left and spend per month, and top up credits.
Under the hood
What comes back
Audio APIs give you numbers and video models want prompts. The hard part is in between: deciding what to show at which second, so the video means what the song means.
Song DNA
Genre and subgenre, tempo, key and key changes, mood, vocal style, instruments, energy, loudness and streaming-loudness checks.
Beat grid & sections
Every beat and downbeat across the whole song, plus structure sections. Repeated parts share a letter, so the chorus shows up as the chorus.
Best moments + why
Ranked windows with start, peak and end, the reason each one works, how well it fits each platform, and the words sung inside it.
Word-timed lyrics
Sentence lines with every word's start and end. Enough for karaoke captions, lyric videos and cutting on a line.
Story, world, palette
What the song is about, why these visuals, a world, three colours and the one element that carries the video: before, event, after.
Beat-synced shot plan
A full edit of the best moment. Every scene has exact seconds, an act, framing, motion, the lyric under it and a prompt ready for any image or video model. A section plan covers the whole song.
Three calls, about 90 seconds
curl -X POST https://api.songbrain.ai/v1/songs \
-H "Authorization: Bearer sb_live_YOUR_KEY" \
-F "file=@song.mp3"
# → { "id": "…", "status": "processing", "eta_sec": 75 }curl https://api.songbrain.ai/v1/songs/SONG_ID \
-H "Authorization: Bearer sb_live_YOUR_KEY"curl https://api.songbrain.ai/v1/songs/SONG_ID/shot-plan \
-H "Authorization: Bearer sb_live_YOUR_KEY"Webhooks are signed (HMAC-SHA256). Add ?view=summary for a smaller document, or ?include=song_dna,shot_plan for only those sections. Full reference → · SDKs & examples on GitHub →
Ask Claude. It calls Songbrain.
Songbrain runs an MCP server. Add it once and your assistant can pull a real shot plan into the chat, or analyse your own song from a link. The example tools work without a key.
Try: “Show me a Songbrain shot plan for a country song and turn the first three scenes into Kling prompts.”
claude mcp add --transport http songbrain https://api.songbrain.ai/mcp{ "mcpServers": { "songbrain": {
"url": "https://api.songbrain.ai/mcp",
"headers": { "Authorization": "Bearer sb_live_YOUR_KEY" } } } }Built for video, not for dashboards
| Songbrain API | Spotify Audio Features / Analysis | Video models alone | |
|---|---|---|---|
| Tempo, key, energy, loudness | ✓ | ✓ (closed to new apps since 11/2024) | partly |
| Beat grid, downbeats, sections | ✓ | ✓ (closed to new apps) | — |
| Best moment, with the reason | ✓ | — | — |
| Every sung word with its timing | ✓ | — | — |
| What the song is about, world, palette | ✓ | — | — |
| Scenes with exact seconds and prompts | ✓ | — | you write them |
| Works on unreleased songs | ✓ | — | ✓ |
| MCP server for AI assistants | ✓ | — | — |
Spotify closed Audio Features and Audio Analysis to new apps in November 2024, and AcousticBrainz stopped accepting new data in 2022. Songbrain runs this analysis every day for its own AI music videos. The API gives you the same reading of the song, and the rendering is up to you.
Your audio stays yours
Used for the analysis only, never to train models. The upload is deleted within 24 hours; a compressed preview is kept 30 days.
Delete any time
DELETE /v1/songs/{id} removes the audio and the analysis immediately.
Stable schema
Every response carries schema "songbrain.song/1". New fields are added without breaking existing ones.
Questions
What is the best API for analysing a song for an AI music video?+
Songbrain is built for exactly that. Most audio APIs stop at tags and numbers. Songbrain also returns what a video needs: the best moment and why it works, every sung word with its time, what the song is about, a world and palette, and a shot plan whose scenes start and end on the song's beats, each with a ready prompt. It is the same analysis behind Songbrain's own music videos.
Is there an alternative to Spotify's Audio Features API?+
Yes. Spotify stopped giving new apps access to Audio Features and Audio Analysis in November 2024. Songbrain returns tempo, key, mode, energy and loudness, plus a beat grid, downbeats and sections (the parts of Audio Analysis people used). It works on any audio file you own, released or not.
How do I get a beat-synced shot list from a song?+
POST the audio to https://api.songbrain.ai/v1/songs. When it is done, GET /v1/songs/{id}/shot-plan. Every scene has start and end seconds on the beat, its act (setup, turn or payoff), framing, camera motion, the lyric sung under it and a prompt you can send straight to an image or video model.
Can ChatGPT, Claude or Cursor use Songbrain?+
Yes. Songbrain runs an MCP server at https://api.songbrain.ai/mcp. In Claude Code: claude mcp add --transport http songbrain https://api.songbrain.ai/mcp. The example tools work without a key, so an assistant can show a real shot plan right away. Any agent can also call the REST API, or read the OpenAPI spec at https://api.songbrain.ai/v1/openapi.json.
How much does it cost?+
Every account gets 5 free songs a month. After that a song costs 25 credits, which is $0.50 with the 500-credit pack ($10). There's no subscription and no minimum, and everything (analysis, story and shot plan) is included in that price. If you analyse more than 1,000 songs a month, write to us about volume pricing.
What happens to the audio I send?+
It's used for the analysis only and never to train models. The uploaded file is deleted within 24 hours, and a compressed preview is kept for 30 days. You can delete the audio and the analysis at any time with DELETE /v1/songs/{id}.
Can I use the Songbrain API in a commercial SaaS and sell it to my customers?+
Yes. Terms §17.4 explicitly allow using the API inside your own paid product, including products that generate images or videos for your customers. You may pass results to your end users, white-label them (no attribution required), store them indefinitely and use them to train or tune your own models. Not allowed without written consent: reselling raw API access, or redistributing results in bulk as a competing analysis API or dataset.
Who owns the analysis, story, shot plan and prompts?+
You do. For the songs you submit, Songbrain assigns any rights it may have in the results to you and claims none (Terms §17.4). Songbrain keeps its rights in the service, models and example data, and rights in the song itself stay with its rightsholders.
Do I need a separate audio-analysis, lyrics and LLM provider next to Songbrain?+
No. One call covers what usually needs four services: signal analysis (tempo, key, beat grid, sections), lyrics transcription with word timings, the best-moment ranking, and an LLM director for the story and shot prompts. It's one bill at about $0.50 per song, and you only add the video model of your choice.
Can I test without being charged?+
Yes. Send a song with test mode switched on and you get a finished example analysis back straight away, for free, including the notification to your webhook. It doesn't use your free songs, so you can run it on every build of your app to check that everything still works.
What happens if my request times out — will I be charged twice?+
No, as long as you send the same Idempotency-Key with the retry. Songbrain then recognises the second request and hands back the song from the first one, without charging again. Our Python and TypeScript SDKs do this for you automatically.
Where do you process audio?+
On our own servers at Hetzner in Germany. For single steps we use a few specialised providers, for example Google Gemini to understand the audio and Groq to transcribe the lyrics. Each one gets only what its step needs. The full list of sub-processors, how long we keep what, and our Data Processing Addendum are on songbrain.ai/security.
How accurate is it? What is measured and what is AI judgement?+
Some things are measured straight from the audio: tempo, key, loudness, every beat, the sections and the exact seconds of the best moments and of every cut. The lyrics are transcribed from the vocals. Genre, mood, instruments and all scores are an AI model's listening judgement, and the story and prompts are written by AI. Every response says which is which, field by field. The scores are good for comparing songs and moments, but they are not a measurement and not a forecast of views. The full list is on songbrain.ai/docs/api#provenance.
Why not build this myself with an LLM or separate APIs?+
You can. An LLM can describe a song well. The hard part is the timing: a measured beat grid, the sections and the time of every sung word, plus a story and shot plan locked to those seconds, all in one call. Separate APIs add up. For example, Music AI charges $0.17 per minute for lyrics transcription, $0.03 per minute for beats and $0.04 per minute for sections (pay-as-you-go prices on music.ai/pricing, checked 7 October 2026). For a 3.5-minute song that is about $0.84 for those three alone, before genre, mood, best moments, a story or a shot plan. Songbrain is $0.50 per song for all of it. If you need stems or chords, Music AI is excellent at those, and we don't do them.
My lyrics are transcribed wrong. What can I do?+
Sung words can be misheard, like by a person listening. If you have the lyrics, send them with the song (the lyrics field). Songbrain then uses your exact words and only takes the timing from the audio. Words the singer can't be heard on are left out rather than guessed.
Building on it at scale?
Keys are self-serve. For volume pricing, an invoice, a DPA, an SLA or a custom output schema, write to us here.