Tempolor v4.8 is Tempolor's current flagship music-generation model. It continues the hierarchical, progressive architecture, producing 48kHz stereo high-quality music. Music generation is decomposed into learning and generation tasks at different levels, achieving coarse-to-fine structured generation: the high level handles musicality and structural organization, the middle level handles semantics and content expression, and the low level handles acoustic detail and high-fidelity reconstruction — balancing musicality and audio quality.

This release aims to take AI music from "usable" to "genuinely good": tracks now run up to 360 seconds; main-melody lines are shaped more strongly and move more naturally; the dynamic rise and fall and section-by-section build-up of real performances replace the flat, monotonous feel of AI music; and overall generation performance is up, with the first section streaming in 15 seconds and a full song finished in under 66 seconds.

In terms of controllability, instruction following across instrument timbre, harmony arrangement and vocal performance is substantially improved, alongside re-creation features such as song Cover, precise Remix and fine-grained audio editing.

Tempolor v4.8 is a continued iteration of the fourth-generation hierarchical, progressive architecture, with a clear lift in overall musicality: vocals and arrangement support each other and blend seamlessly, making each piece more complete and more moving.

Basic generation: better-sounding, more controllable, faster

• Catchier melodies: Stronger shaping of the main melodic line, with more natural melodic movement that sits closer to what the ear finds pleasing.

• Livelier sound: Simulates the dynamic rise and fall and section-by-section build-up of real performances, leaving the flat, monotonous feel of AI music behind for tracks that stay fresh on repeat.

• More precise control: When you adjust instrument timbre, harmony (Backing Vocals) or vocals via tags or prompts, the model responds closer to intent, making the process more predictable and the result more likely to land in one pass — what you imagine is what you get.

• Longer and faster: Tracks up to 360 seconds; with streaming generation the first section is ready in 15 seconds and a full song in under 66 seconds.

• Stable and reliable: Music pass rate and generation success rate both exceed 99.9%, with support for 40 languages including Chinese, Japanese and English.

Song Cover: more accurate melody reconstruction, a full upgrade to Remix

• Melody reconstruction: With identical lyrics, melody reconstruction accuracy exceeds 99%.

• Returning to the motif: Further improved handling of difficult material — cover lyrics that differ in length from the original audio, or irregular song sections — so the model converges back to the main melody quickly even during free re-creation, avoiding loose melodic lines and leaving more dependable room for re-creation.

• Remix rewriting: Reshape existing tracks at the level of style and arrangement, efficiently exploring many musical directions from the same material.

• Arrangement and vocals: Instruction following for mainstream styles is broadly stronger, with markedly less interference from the original track's style; vocals are natural and fluent, and fit closely with the arrangement for a stable, cohesive result.

Based on Chinese and English test sets (30 each, 60 total), compared against Mureka v9, Suno v5.5 and MiniMax V2.6, covering the Meta Audiobox Aesthetics and SongEval evaluation systems.