Opus 5.5 Writes the Theme, Suno Scores the Film: A Chinese Brand TVC Just Ran the Full AI Ad-Music Pipeline

Updated:

On November 6, 2026, AI creator Haixin (Ring Hyacinth) posted a follow-up on X explaining the music workflow behind her latest film. Released the day before at the invitation of Chinese apparel brand Semir, the spot — promoting the season's flagship "Daxing'anling Goose Down" jacket — was made with Opus 5.5 and ByteDance's video model Seedance 2.5. When commenters asked what Opus 5.5 had actually contributed, her answer was: besides early storyboard animation, the film used Opus 5.5 to write a theme melody, then Suno to expand that melody into background music for different scenes, so that "the music across the entire film stays unified" (original post).

Video cover for the Semir × SHAMA TVC "A Gift from Daxing'anling"

The post itself is barely a hundred words and drew under five thousand views, but the workflow it describes deserves serious attention: this is a publicly documented case of a real, aired Chinese brand TVC whose music was handled entirely by an AI toolchain — a large language model composed the theme, and a music generation model did the scene-by-scene expansion. For the ad-music niche, this is the first time the full "AI composition plus AI arrangement" pipeline has run end to end.

Why This Pipeline Works

The core requirement of advertising music is not "sounds good" — it is "stays unified." A TVC typically breaks into a dozen-plus scenes, each with its own mood, tempo, and instrumentation, but the thematic motif must run through the whole film: audiences forget lyrics, they remember melodies. In the traditional workflow, a composer writes the theme and arrangers produce scene-specific variations — essentially "one theme, N rewrites." Semi-structured rewriting like this is exactly what current generative tools handle best.

The toolchain integration is already off the shelf. Suno's official audio upload feature (help documentation) explicitly supports two operations on imported audio: Extend (continue generating after existing material) and Cover (transform the uploaded audio into a new song version) — "take a theme melody and generate different-scene BGM" is precisely a combination of these two functions. Haixin did not disclose in what form Opus 5.5's theme melody was handed to Suno (text notation, MIDI, or rendered audio — none of this is publicly documented, so no claim is made here), but the division of labor itself — LLM defines the theme, generative model does the expansion — has clearly worked.

The video side came together in 2026 as well. Seedance 2.5, released by ByteDance's Seed team at the end of July (official page), doubled single-generation length to 30 seconds at native 1080p and supports up to 50 multimodal reference inputs; Chinese tech media widely framed it as the moment AI video moved from "can generate" to "deliverable." For 15–60 second TVC spots, that means both picture and music can now close the loop inside an AI toolchain.

On the demand side, this fits Semir's own rhythm. This is not the first collaboration between Haixin's team and Semir — in July 2026 they made an AI short film for Semir's new IP "SHAMA." And "Daxing'anling Goose Down" is Semir's flagship line this season: according to a brand press release syndicated by iFeng (October 27, 2026), Semir put the place of origin directly into the line's name, set a "double 50" down-selection standard ("born at 50 degrees north, raised at minus 50 degrees"), and at the end of October released a documentary short with Xinhua News featuring celebrity endorser Li Xueqin. A brand letting AI produce the TVC for its most heavily promoted product line — and saying so publicly — is a signal that matters more than traffic numbers.

Market Context: The Pie Is Being Re-Cut

Advertising and media scoring has long been an underrated market. Research firm Coherent Market Insights estimates the stock music market at roughly $1.5 billion today, growing to about $2.8 billion by 2030, a compound annual growth rate of around 10.9% (a single-firm estimate). On pricing, industry convention puts custom ad music from a few thousand dollars upward, with top projects reaching tens of thousands, while library tracks are an order of magnitude cheaper — on the VI-Control forum, a working commercial composer quoted a custom range of $2,000 to $50,000-plus, depending on media scope and buyout terms.

AI-generated music has near-zero marginal cost, and it strikes directly at the foundation of this pricing structure. A global economic study published by CISAC in October 2024 (summarized by the Ivors Academy) made a stark projection: by 2028, AI-generated music is expected to account for 20% of traditional streaming revenues — and 60% of music libraries' revenues. Music libraries are precisely the main supply side for advertising, film, and game scoring. The same study estimated that without intervention, human creators stand to lose nearly a quarter of their income within four years. Ad music — fast-turnaround, fragmented-budget, function-first — is widely seen as the first segment AI will swallow.

The immediate effect for brands: mid-tier marketing content that could never afford custom scoring now has a "subscription scoring" option. The first comment under Haixin's TVC post is literally "how much did it cost?" — whoever asked had clearly done the math.

Rights and Compliance: Usable Commercially, But Perhaps Not "Ownable"

Brands scoring ads with Suno are not in a legal gray zone — and the landscape is settling fast.

First, platform licensing. Under Suno's official rights policy, songs downloaded while on a paid subscription come with commercial use rights; the free tier is non-commercial only. The barrier to compliant brand use is a clear subscription cost, not a complex licensing negotiation.

Second, copyright ownership — the subtler part. The framework established by the U.S. Copyright Office's AI policy (Part 2 of its report, January 2025) holds that purely AI-generated output is not copyrightable; only the human-authored expressive contributions can be registered. That means the "commercial use rights" a brand gets with AI scoring are not copyright — a competitor could, in principle, lawfully use the same or a similar AI-generated score, and the brand cannot obtain a registrable, exclusive audio asset the way it can with a commissioned custom track. For companies that manage sound logos as long-term brand equity, this is something AI scoring currently cannot deliver. Chinese judicial practice points the same way: in AIGC-related rulings, the Beijing Internet Court has held that rights in AI-generated content can only belong to a natural person or legal entity, premised on original human input — pure machine output is not protected (see Hongan Law Firm's analysis of generative-AI music ownership disputes).

Third, training-data litigation risk is converting into licensing deals. After the major labels sued Suno and Udio in 2024, the situation flipped in late 2025: Warner Music settled with Suno in November 2025 and signed a licensing partnership, and Universal Music settled with Udio in the same period, with a licensed new platform to follow (NPR; Law Commentary's litigation overview). For brands, the 2024 compliance question — "will using Suno drag us into a record-label lawsuit?" — has largely eased by 2026. The cost is being borne by session musicians instead: in June 2026 the American Federation of Musicians sued Universal, Warner, and Atlantic, alleging the labels licensed members' recordings to Suno and Udio as training data without compensation or credit (Music Business Worldwide).

For Professional Ad Composers: Displacement, or Stratification

Put CISAC's 60% music-library projection next to this Semir TVC, and the impact on professional ad composers looks less like wholesale replacement than stratified squeeze.

The squeezed layer is functional delivery: bulk background music, e-commerce clips, feed-ad scoring — jobs where the brief is "fast, cheap, non-infringing," and the AI toolchain already delivers passable answers. What holds is the two ends: the creative top tier (director-level scoring for flagship campaigns, where brands pay a premium for a named composer) and precision-to-picture work. Ad scoring demands frame-accurate sync — music must land on specific cuts and resolve as the logo appears; this hit-point-level control is still approached by generate-select-regenerate loops and requires a human making final adjustments in a DAW (this is the author's analysis based on industry workflow, not an empirically established conclusion). Haixin's own workflow confirms the stratification: Opus 5.5 and Suno handle theme and expansion, while the judgment calls — which version, how to splice, whether the whole film coheres — remain with the creator.

One more layer is easy to miss: what this pipeline changes may not be "who writes the score" but "who can afford a score." Vast amounts of brand content previously went without music — no budget for custom scoring, so brands used free libraries or nothing. Subscription AI scoring drops the floor for "having music at all," so some of the new demand may not be displacing human composers at all (this is the author's inference, pending more deployment data).

The Counter-Case: A Working Pipeline Is Not a Mandate to Floor It

Beyond the efficiency story, brands should weigh at least three limiting perspectives.

First, listener attitudes are worsening, not improving. Research by data firm Luminate (NPR report) compared attitudes from May to November 2025 and found net sentiment toward AI music sliding from -13% to -20%, with the steepest declines among Gen Z and Gen Alpha; about a third of listeners remain indifferent, but the number switching from positive to negative is growing. For advertisers this creates a disclosure dilemma: not disclosing AI scoring invites passive reputational risk; disclosing it invites reflexive backlash from part of the audience. R&B singer SZA told i-D she feels "at war" with AI — the chain from artist stance to brand pressure, transmitted through fandoms, is real.

Second, supply glut is not proof of demand. The same NPR report cites Deezer's detection data: about 44% of daily uploads to the platform are AI-generated, yet AI tracks account for under 3% of total streams — and a majority of those streams were judged fraudulent. AI music is exploding on the production side without earning an equivalent place on the listening side. Ad music is functional consumption, so the mismatch doesn't map directly — but it is a reminder that supply-side abundance is not evidence of quality.

Third, positive AI-music cases exist, but under demanding conditions. Xania Monet, a self-disclosed AI project, has charted on Billboard and signed a multimillion-dollar deal with Hallwood Media in fall 2025 — but that rested on a full operation of persona, narrative, and transparency, not generation quality alone. A brand wanting a positive AI-scoring story needs far more than a subscription.

Conclusion: After the Pipeline, the Divide Is Human

The significance of the Semir TVC is not that it dazzles, but that it turns something into a "case": the full AI workflow for commercial music — theme writing, scene expansion, aired deliverable — now has usable tools, a compliance path, and a reference precedent at every step. CISAC's 60% music-library projection won't happen overnight, but the direction is clear: functional scoring capacity will keep shifting into the toolchain.

The real divide is not the tools; it is people. The most valuable moves in this pipeline remain the judgments — what the theme is, how unified is unified enough, which take is the right one. The Haixins of the industry have gone from executors to curators. For professional composers, what shrinks is functional orders; what expands is "the taste to direct AI." For brands, only the notes got cheaper — exclusivity over sound assets and listeners' trust still price the old way.

Sources