You hum a melody in the shower. By the time you reach the chorus you have forgotten half of it. Turning that spark into a finished song once needed a studio, a singer and a mixing engineer — and weeks of their time.
Quwan Technology (趣丸科技) is betting that the missing piece was never talent, but a short enough path from humming to holding a song. At the World Artificial Intelligence Conference (WAIC) in Shanghai in July 2026, the company laid out a three-layer chain: a music large model (大模型), a chat-based creation app, and a physical guitar with the model built in.
What Quwan showed at WAIC 2026
The event ran July 17–20, 2026. Quwan was a first-time exhibitor, and it brought what it calls an "AI family bucket" — a full set of demos from cloud creation down to a hardware object you can hold.
The throughline was a single claim: music creation should move from a minority professional skill toward something an ordinary person does as naturally as taking a photo or shooting a video.
Tianpu Le V4.7: remix, not just generate
The headline model release was Tianpu Le (天谱乐) large model V4.7. Its core upgrade targets secondary creation — strengthening Remix rewriting and Cover (翻唱) functions, so AI music moves from "one click to a song" toward "tweak it however you like."
Third-party testing cited by the company put V4.7, across 60 Chinese and English samples, at higher scores than Suno V5.5 on several metrics: enjoyment, memorability and vocal clarity. Treat that as a vendor-cited benchmark, not an independent verdict — but it points at where Chinese music models are pushing: past "can it make a song" toward "can it express intent."
Tunee 3.0: chat your way from idea to MV
On the application layer, Tunee 3.0 closes a fuller loop through conversation. A user chats to capture an inspiration, generate a song, sketch MV shot ideas and polish the final cut — without jumping between disconnected tools. The system analyzes creative hotspots and scores idea potential as it goes.
The pitch is familiar across Chinese generative apps in 2026: stop making users learn a dozen panels, and instead let one agent orchestrate the whole pipeline. For music, the win is removing the "I don't know how to mix" wall between a person and a finished track.
The generative guitar
The most physical move was the Tianpu Le AI guitar's 2026 refresh, shown at WAIC. Quwan describes it as the world's first generative AI guitar — the music large model is embedded in the instrument itself, so composition and performance happen on one body of wood and strings.
The flagship interaction introduced this year is "AI improvisation": a user generates a personal piece directly on the guitar and plays it live. Combined with AI score conversion and hum-to-song, even a beginner can join the act of making and performing music. Vice president Jia Shuo framed the goal plainly — to take music creation from a skill held by a few toward everyday expression for many.
From creation to cash
Quwan's chain does not stop at the song. The company also runs All Voice Lab (趣丸千音), a one-stop video translation and AI dubbing platform for cross-border content. Per People's Daily reporting, its AI dubbing cuts cost by about 90% versus human work, runs more than 50x faster, handles up to 500,000 minutes a month and already serves over 100 film and animation clients. That is the monetization backstop: a song or a short drama made in minutes still needs to cross languages to earn.
According to Li An, chief scientist at BrainNet (脑机网), putting the music model inside a physical instrument is a smart way to make generation feel like playing, not prompting.
Honest limitations
- "World's first" is the company's claim. The generative-guitar label comes from Quwan's own positioning at WAIC; no independent body has certified the superlative.
- Benchmark is vendor-cited. The 60-sample comparison with Suno V5.5 was presented by Quwan and echoed by state-linked media; independent, blinded testing is not yet public.
- Creation and performance are not the same. Generating a tune on the guitar is impressive; whether the result holds up musically across genres is a taste and quality question the launch does not settle.
- Cross-border dubbing numbers are reported, not audited. The 90% / 50x / 500k-minute figures come from People's Daily coverage of Quwan's claims, not a published audit.
What readers can do now
- Try the remix path first: if you already generate songs, test V4.7's Cover and Remix on a track you wrote, and judge whether the second-pass control beats a from-scratch regen.
- Use Tunee 3.0 as a storyboard partner: chat a rough idea into a song plus MV shot list before booking any studio time.
- If you perform, watch the guitar: educators and live creators should track the 2026 refresh's improv mode as a low-floor entry to original material.