Gege AI launches billion-parameter Mandarin music model with ByteDance revenue-sharing deal

Gege AI has launched a one-billion-parameter music generation model trained entirely from scratch, purpose-built to solve the unique challenges of generating Chinese-language songs.

The end-to-end model takes lyrics, style descriptions, and vocal preferences as input and generates a complete 3-minute stereo song with both vocals and accompaniment, outputting separate vocal and instrumental tracks.

Rather than fine-tuning existing models, Gege AI chose to train from scratch because mainstream AI music models are designed for English. Mandarin presents fundamentally different challenges: every character has a fixed tonal contour, syllable boundaries are crisp and unambiguous, and the alignment between lyrics and rhythm is far more constrained than in English.

The company built a three-layer technical stack: a dual-stream architecture where vocals and accompaniment generate in parallel with cross-stream attention for real-time alignment; phoneme timing soft-alignment that maps each character's position on a timeline before generation; and a layered multi-dimensional conditioning system (AdaLN-Zero) for controlling emotion and genre throughout the generation process.

On the business side, Gege AI signed a copyright revenue-sharing agreement with ByteDance. Songs generated through Gege AI can be legally uploaded to ByteDance's ecosystem including Douyin, CapCut, Qishui Music, Xigua Video, and Toutiao. Revenue from membership fees, digital albums, and ad sharing will be distributed per contract terms.

Gege AI also announced the launch of a Chinese folk music AI model project. The team plans to travel across China to record authentic performances of traditional instruments including Qin opera, Pingtan pipa, Hulusi, and Suona from veteran folk artists, using these recordings to train a model that can genuinely capture the essence of Chinese traditional music.

The dual focus on technical architecture and monetization channel positions Gege AI's approach as a potential blueprint for sustainable AI music businesses.