ElevenLabs' Instant Soundtracks scores your video and times the effects. Library composers feel it first online

ElevenLabs launched Instant Soundtracks on 15 Oct 2026, an update to its ElevenCreative suite. Upload a video and the model writes an original score for it. It also watches the footage for moments that call for a sound effect and drops each one at the matching timestamp; the company's example is a whoosh on the logo at the end of an ad. ElevenLabs says the model is tuned for short-form video, with tighter timing around cuts, transitions and on-screen action.

Generating music from a prompt is not new, and ElevenLabs has done it since Eleven Music launched in August 2025. What is new is that the tool now does the editor's job too: it watches the picture, hits the cuts and places the effects. That is the work a video editor used to hire a library track and a sound-effects pack for.

What does it actually do?

Three things, according to the announcement:

• Scores the video. The music can change mood between scenes. ElevenLabs gives the example of EDM or R&B under a fast scene moving to acoustic guitar under a slower one.
• Places sound effects. In the Music Video Editor the effects sit on their own layer, separate from the soundtrack. You can add effects, move them, swap the music or fix one moment without regenerating the whole thing.
• Uses your audio as a reference. If the clip already has sound, the model can read its genre, crescendos, BPM and instrumentation and shape the new score to match.

For developers, the Video to Music API exposes the same features as flags: add_sfx for effects cues, separate_tracks to return music and effects as separate files in a ZIP, and use_audio_as_reference. One request takes up to 10 videos, 200MB combined and 600 seconds in total. There is also a sign_with_c2pa option that signs MP3 output with C2PA provenance data, a useful step for anyone who wants to show a track is AI-generated.

The announcement gives no pricing, no word on training data and no performance figures.

Who can use it, and for what?

This is the part the launch post leaves out, and it decides who feels the impact. The Eleven Music terms, last updated on 9 Oct 2026, set out rights by plan. Every plan from Free through Business, plus Enterprise Music Lite, allows "all online and offline commercial use" except film, TV, radio and Studio Games. Only the full Enterprise Music plan removes those exceptions. Free users must credit Eleven Music when they distribute; paid plans need no attribution. The games exclusion covers "Studio Games"; the terms treat indie games, defined by revenue thresholds, separately.

So a creator on a self-serve plan can score a YouTube video, a podcast trailer, a product ad for social or an indie game. A production company scoring a TV series or a radio campaign has to sign an enterprise deal first.

Who loses?

The obvious answer is production-music and library composers, and the industry's own numbers back that up. A study by PMP Strategy for CISAC, the global body for authors' societies, projected in December 2024 that by 2028 generative AI music would account for around 60% of music libraries' revenues, against about 20% of streaming platforms' revenues. It also put 24% of music creators' revenues at risk by then. Those are forecasts, not measurements, but the gap is telling: libraries were already expected to be hit three times harder than streaming.

Instant Soundtracks makes the library use case more direct. A library cue is valuable because it is ready to cut to picture. A tool that scores to picture itself skips the search, the edit and the licence. Sound-effects recordists are in the same position. The Professional Sound Alliance we covered in September is campaigning against libraries being scraped into AI models without permission or payment. A model that places effects on its own timeline raises the same question.

The squeeze is not new. Artlist, a stock-music subscription service, added Google's Lyria 3 Pro in March 2026 so its users can generate tracks of up to three minutes, and MBW reported that Google positions the model as a potential alternative to royalty-free production music services.

Who might gain?

Video editors and small creators get faster, cheaper, cleared audio, and separate stems they can mix. Developers building video tools get scoring as an API call. The rights holders who licensed Eleven Music's data might gain too: MBW reported at launch that ElevenLabs had deals with Merlin and Kobalt, with Kobalt talking about AI revenue streams for songwriters who choose to take part. No payment terms were disclosed then, and we have not seen any since.

Libraries are taking different routes. Epidemic Sound launched Adapt in September 2025, which uses AI to edit its own human-made catalogue to fit a scene, with an opt-out for artists, a $1M pool for AI-related adaptation and a 43% rise in overall artist bonus payouts in 2026. That model keeps the composer in the loop and pays them for it.

What we think

The licence tiers tell you where the first damage lands. Film, TV and radio, the most visible and best-paid sync work, are fenced off behind Enterprise Music. ElevenLabs doesn't say why, and nothing in the terms suggests the fence is there to protect composers. What is open to everyone is online video: social clips, ads, explainers, podcasts, corporate content. Our reading is that this is the high-volume, low-fee end of the production-music business, and we think the composers who rely on it, not the ones writing cues for network TV, will feel this first.

Editable layers and an API make it worse, not better, for human cues. A finished library track had one advantage over AI output: it sounded finished. Separate stems and timed effects you can tweak close that gap. Through the API, scoring can sit inside every editing app, and the editor never browses a library at all. Fewer searches means fewer chances for a human cue to be picked.

The fair version exists, and ElevenLabs could build it. It already has a rights story. Our coverage of its UMG deal found that, under ElevenMusic's terms, the licence an artist grants does not by itself entitle them to any payment. A scoring tool that replaces library cues should pay into a pool for the composers and recordists whose work trained it, the way Epidemic Sound pays when its own tracks are adapted. Until ElevenLabs publishes who is paid and how, we would treat "cleared for commercial use" as a promise to the buyer, not to the musician.

For library composers, the practical advice is unglamorous: check which of your libraries license catalogue for AI training and on what terms. Push for opt-outs and per-use pay. And look at where your income actually comes from. Broadcast cues and bespoke scoring sit behind the enterprise fence for now; anonymous online background music does not. That is the same replacement risk musicians' unions are now taking to the bargaining table.