Masters of Taste: Can AI Turn a Videographer Into a Film Composer?
- Theoplis Stewart II
- Aug 9
- 10 min read
The old version of this story begins with a filmmaker waiting on a composer. The picture is nearly locked. The client wants a final cut tomorrow. There is no budget for an orchestra, barely a budget for stock music, and every library track that almost works seems to collapse at the exact moment the video needs to rise.
The new version begins with a text box.
A videographer can now describe a cue in plain language — restrained piano, low strings, no triumphant brass, eighty beats per minute, tension building after twenty seconds, a clean emotional lift under the final line of narration — and hear a finished piece of music minutes later. With tools such as Eleven Music, the person cutting the film can generate, revise, extend, inpaint, restructure, and in some workflows separate music into stems without ever writing a note on staff paper.
This is more than another AI novelty. It is the beginning of a serious rearrangement of who gets to make a score, what it costs to make one, and which skills matter when technical access is no longer the main barrier.
The score is becoming a conversation
ElevenLabs describes Music v2 as a model capable of handling full songs, section-by-section composition, improved arrangement, audio-reference guidance, and local regeneration of selected passages. In other words, the workflow is moving away from one prompt producing one untouchable blob of audio. A creator can increasingly say: keep the opening, rewrite the middle, remove the choir, make the transition less heroic, preserve the percussion, and land the final chord two seconds later.
For small video production, that distinction is enormous. Stock libraries solved the problem of access to music, but they never solved the problem of exact fit. Generative music attacks fit directly. The soundtrack can be made for the edit instead of the edit being bent around the soundtrack.
ElevenLabs is explicit about the market it is pursuing: film, television, advertising, gaming, podcasts, branded content, and social video. Its newer products are designed to let a creative team brief the system in the language of mood, tempo, structure, instrumentation, and brand identity, then export music for commercial production under the terms of the relevant plan.
Can a non-composer make something that sounds like a major film score?
Yes — and that answer needs an immediate qualification.
A person with no formal composition training can already generate music with the surface vocabulary of expensive cinematic scoring: hybrid orchestra, pulsing low strings, intimate felt piano, processed percussion, trailer impacts, evolving synth beds, choral textures, harmonic swells, and polished mastering. On a thirty-second commercial, a ninety-second documentary sequence, a social campaign, a wedding film, a nonprofit profile, or a short branded piece, many viewers may have no practical way to know whether that cue came from a composer, a library, or a model.
But sounding cinematic is not the same thing as scoring a feature film at the level of the greats. Elite film composition is not merely the production of beautiful audio. It is dramatic architecture across time. It involves spotting the film with a director, deciding where music should not exist, developing themes across ninety or one hundred and fifty minutes, reshaping harmony around dialogue, tracking character psychology, handling revisions after picture changes, orchestrating for specific players, communicating with editors and mixers, and preserving a musical identity across dozens of cues.
That is why asking whether an AI cue can “beat Hans Zimmer” or another famous composer is the wrong test. A model can imitate the sonic grammar of prestige scoring with astonishing speed. The harder question is whether a human director of that model can sustain dramatic intention, restraint, continuity, and surprise across an entire film.
The surface is becoming cheap. Architecture is still expensive.
The most important shift may be that technical fluency and musical judgment are separating. A creator no longer needs years of keyboard technique to audition an orchestral idea. But the system still needs someone to know that the first version is too sentimental, the second is too busy under dialogue, the third resolves too early, and the fourth finally makes the cut feel inevitable.
That person is not necessarily a traditional composer. They may be a director, editor, producer, photographer, or videographer. Their instrument is increasingly selection.
This is where the phrase “master of taste” becomes useful. When creation becomes abundant, the scarce skill is not generating more. It is recognizing what belongs.
Where AI can genuinely beat the old production model
I could not find a defensible professional study showing a prompt-only AI score defeating an elite feature-film composer in a controlled head-to-head evaluation. Claims that generative music has already surpassed the best living composers usually confuse polish with authorship or automated benchmarks with human dramatic judgment.
There are, however, narrower areas where AI is already winning. Research systems such as FilmComposer have reported state-of-the-art results among automated systems for musicality, video consistency, diversity, and musical development by combining visual analysis, rhythm control, generative models, arrangement, and mixing agents. VidTune, a 2026 research system built around generative soundtrack selection, found that creators benefited from being able to compare musical options in context and refine them through natural-language edits.
Neither result means a machine has out-composed John Williams. It means the machinery of matching music to moving images is becoming more intelligent.
For a small production, that can be more important than abstract greatness. A bespoke human score may require a brief, a spotting conversation, composition time, revisions, arrangement, recording or sample programming, mixing, delivery, and licensing paperwork. A generative workflow can audition multiple custom directions during the edit itself. ElevenLabs even markets its current creative product around eliminating conventional sync-fee and clearance delays for eligible generated material. That is not the same artistic service, but it is a radically different production equation.
The videographer becomes a music director
The weak way to use AI music is to type “epic cinematic score,” accept the first result, drag it under the picture, and call the job finished. The professional way looks much closer to directing a department.
The best prompting guidance from current music systems is already drifting toward the language working composers and producers use: genre, instrumentation, tempo, structure, emotional trajectory, density, dynamics, use case, and explicit exclusions. ElevenLabs recommends building section by section and iterating rather than treating generation as a one-click event.
A strong creator also works from picture outward. Before asking for music, identify the cue’s job. Is the scene asking for suspense, grief, propulsion, irony, warmth, scale, or simply continuity? Where is the dramatic turn? Which line of dialogue must remain exposed? Does the music need to announce a transition or disappear before it? A model can generate sound. It cannot take responsibility for the storytelling decision unless the human first makes that decision legible.
What the artists are saying: tool, threat, or both?
The music world is not divided cleanly into believers and resisters. The more interesting line runs through the middle: artists willing to experiment while insisting that authorship, consent, identity, and human intention remain visible.
ElevenLabs’ 2026 Eleven Album offered one of the clearest examples of that collaborative argument. Liza Minnelli framed her participation around “connection and emotional truth” and said the new tools mattered when used in service of expression rather than as a substitute for it. Art Garfunkel’s formulation was even more concise: “The human remains at the center.” Brazilian producer and director KondZilla said the quality of the system impressed him while describing it as a way to translate his existing aesthetic into a different sonic context.
Grimes has taken the experiment further, allowing others to work with an AI version of her voice and treating identity itself as something that can be licensed and collaboratively extended. But even she has acknowledged that musicians’ economic fears are not imaginary, especially for session players and in a streaming environment where cheap, anonymous music can flood playlists.
Nick Cave represents the opposite anxiety with unusual clarity. Asked directly in April 2026 about AI music generators, he did not dismiss their quality. His fear was almost the reverse: that they could become technically excellent — even superior in presentation — while stripping away the human struggle he believes gives music spiritual meaning.
That concern sits beside a broader rights revolt. More than 200 artists, including Billie Eilish, Stevie Wonder, Miranda Lambert, Nicki Minaj, and Katy Perry, signed an Artist Rights Alliance letter warning against AI practices that exploit artists’ work without permission or compensation. The argument is not that technology must stop. It is that the people whose work gives the technology value should not disappear from the economic equation.
Teachers are already building a discipline for the Wild West
If the technology makes music generation easy, education has to make judgment difficult again.
Berklee College of Music’s current AI principles explicitly center human experience, artist rights, practical technological fluency, and adaptability. That framing matters because a serious curriculum cannot simply teach students how to get better outputs. It has to teach them why one output is better than another, what it borrows from, whether it serves the scene, and what ethical or legal baggage travels with it.
Music educators are beginning to formalize this. Marina McLerran, writing in Music Educators Journal in 2026, describes generative systems as possible composition assistants, exploratory tools, multimedia-production aids, and vehicles for self-expression — not replacements for foundational musical understanding. A classroom study of Suno in secondary music education used a simple iterative model: Prompt, Generate, Review, Regenerate, Finalize. The important word is review.
At the 2026 International Society for Music Education conference, another AI-integrated composition model pushed even further beyond one-click creation. Its four phases were: Ideate and Prompt; create lyrics, melody, and structure with human leadership; Audit and Explain the model’s output for coherence, bias, and provenance; then Produce and Refine through arrangement, mixing, mastering, and version control.
That is beginning to look less like cheating and more like a new form of apprenticeship. The student is not judged for whether the machine can make a competent track. The student is judged for the quality of the brief, the sophistication of the critique, the decisions made during revision, and the ability to explain why the final music belongs with the image.
A professional AI-scoring workflow for someone who cannot write music
Spot the picture before you generate. Mark the emotional turns, dialogue windows, reveals, transitions, and exact cue length. Do not ask for music until you know what the music is supposed to do.
Write a cue brief, not a vibe word. Define tempo range, instrumentation, density, structure, emotional arc, reference qualities, and what must be excluded. “Sad” is weak direction. “Sparse felt piano and low cello, no percussion for the first twenty seconds, restrained harmonic tension, subtle lift under the final sentence, unresolved ending” is usable direction.
Generate alternatives. Professional taste is comparative. Audition several approaches against the actual edit before deciding which musical language belongs to the story.
Edit the music instead of worshipping the generation. Rebuild sections, trim phrases, extend sustains, regenerate weak transitions, separate stems when available, and reshape the cue around picture changes.
Mix for the video, not for the song. A gorgeous full-range track can be terrible under narration. Make room for speech, sound effects, silence, and location audio. The best score often wins by knowing when to recede.
Bring in human musicians when the cue exposes the model’s weakness. A live soloist, vocalist, percussionist, or instrumental overdub can transform a competent generation into a specific performance.
Document the tool, plan, license, references, and human edits. The more commercial the production becomes, the more provenance matters.
Copyright changes the meaning of authorship
There is also a legal reason not to stop at the prompt. The U.S. Copyright Office has said that purely generative output is protectable only to the extent a human author determines sufficient expressive elements. Human-authored material carried into the output, creative selection and arrangement, or meaningful modifications can support copyright protection. Merely supplying prompts, by itself, is not enough.
For the videographer, that creates an incentive to act like an editor and producer rather than a customer at a vending machine. The more you shape structure, combine material, perform, arrange, and make expressive decisions, the stronger the claim that there is human authorship in the finished work.
Platform choice matters too. ElevenLabs says its current music model is trained on licensed material and cleared for broad commercial use under its applicable terms. The rest of the market is more complicated. Suno and Udio have faced major copyright litigation and licensing disputes, and a German court ruled against Suno in a GEMA case in July 2026. A creator cannot treat every “AI music” button as legally interchangeable.
Masters of taste. Masters of dissemination. Masters of collaboration.
The age of generative media is producing a strange inversion. When almost anyone can generate polished material, polish itself stops proving much.
The first new master is the master of taste: the person who can hear ten technically competent cues and know which one turns the scene from manipulative to moving.
The second is the master of dissemination: the person who understands audience, platform, pacing, packaging, trust, and context. Cheap creation does not create attention. A 2026 study of AI music on streaming platforms found that the overwhelming majority of identified AI tracks received few or no plays. Infinite supply makes distribution and identity more important, not less.
The third is the master of collaboration: the person who knows when the model is enough, when the musician is necessary, when the editor should reshape the cue, when the sound designer should take over, and when silence is stronger than another layer of orchestration.
This may be the most consequential implication for small video production. AI does not merely remove a specialist from the budget. Used well, it gives the filmmaker access to the language of that specialty — and therefore a new responsibility to learn how to direct it.
So, can they do it at that level?
For a short film, commercial, documentary scene, creator video, branded piece, or social campaign, a non-composer with exceptional taste can already produce music that sounds far more expensive than the production that contains it. In some contexts, the result may be functionally better than hiring a mediocre composer or forcing a generic library track into the edit, because the generated score can be designed around the exact emotional and temporal shape of the picture.
That still does not make the user Hans Zimmer, Hildur Guðnadóttir, John Williams, Ludwig Göransson, or Trent Reznor and Atticus Ross. Great film scoring is not a preset called “cinematic.” It is sustained judgment under narrative pressure.
But the threshold has changed. The question is no longer whether someone who cannot read music can make a convincing score. They can. The question is whether they can develop enough taste, musical vocabulary, editorial discipline, and collaborative judgment to make that score mean something.
AI has not made everyone a composer. It has made musical decision-making unavoidable.
Sources and further reading
AI disclosure: This article was researched and drafted with AI assistance under human editorial direction. Sources were reviewed and linked to distinguish documented claims from analysis.
Comments