TianPuLe Large Model Releases V4.7 with Enhanced AI Music Remixing and Cover Capabilities

During the 2026 World AI Conference (WAIC) in Shanghai, Quwan Technology officially unveiled TianPuLe Large Model V4.7, integrating it into its conversational music creation agent Tunee and opening it to developers and enterprise users via the TianPuLe OpenAPI.

Unlike earlier updates that focused on audio quality or song length, V4.7 prioritizes creative control and iterative editing. The industry shift is clear: after solving 'can AI generate a song,' the question now is 'can users keep editing that song until they are satisfied.'

The centerpiece of V4.7 is its upgraded Remix and Cover capabilities. Users can upload existing music material and redesign its genre, arrangement, or vocal style while preserving the core melody. A抒情 (lyrical) melody can be reimagined as electronic, rock, or light jazz.

For cover song generation, where new lyrics often differ in syllable count and structure, V4.7 introduces a 'rapid motif regression' technique: the model rewrites around new lyrics but returns to the main melody at key structural points, maintaining the original song's recognizability.

V4.7 also addresses reference audio leakage. When learning a song's melody and structure, the model minimizes direct copying of original vocals and timbre, producing cleaner, more independent outputs.

On basic generation quality, V4.7 optimizes melodic motifs, phrasing breath points, and song dynamics to reduce the mechanical repetition common in AI music. Spatial separation between drums, bass, harmony, and vocals has also improved, particularly in slow-tempo and relaxed genres.

In third-party tests across 60 Chinese and English samples, V4.7 was benchmarked against Suno V5.5, Mureka V9, and MiniMax V2.6. It scored highly on content enjoyment, memorability, and audio clarity under the Meta Audiobox Aesthetics and SongEval evaluation frameworks.

The TianPuLe team redesigned their self-supervised music representation model to decouple timbre, melody, and rhythm, then recombine them according to user prompts for finer creative control.

Overall performance improved by approximately 20%, with full song generation completing in under 60 seconds at 48kHz stereo output. As the barrier to generating a song continues to fall, V4.7 represents the next step toward integrating AI music into real creative workflows.