YUE2 music model takes LoRAs well — dev offers free fine-tunes on a 5090
-becausereasons- · reddit · 2026-09-19
A Reddit user reports YUE2 is the first frontier-level music model that supports LoRA fine-tuning well, having trained four niche-style LoRAs so far: militant roots reggae, Sanremo-era Italian pop, French chanson (with a spoken-word trick), and a Bulgarian choir + Tuvan throat-singing set in progress — demos and weights already on Hugging Face.
He's offering free community fine-tunes: you bring the idea and dataset, he brings the RTX 5090, captioning pipeline, and release.
- Dataset requirements: 15-30 tracks in one coherent style, each under 5 minutes, FLAC or high-bitrate MP3, no live bootlegs; plus lyric language and a 2-3 sentence target-sound description
- Styles he'd love: Ethio-jazz, Tuareg desert blues, Anatolian psych, Russian bards, fado, Qawwali, 70s Bollywood disco, enka/city pop, Tropicália, Klezmer, Balkan brass, and more
- Not interested in: minimal techno, death metal, country, generic EDM/lo-fi
More from Multimodal
- Experimental video mixes visual styles, generated with Seedance 2.5 on Adobe Firefly — umesh_ai · 2026-11-04
- Hailuo AI claims H3 scored perfect on language tests; Japanese creator praises its Japanese signage fidelity — Hailuo_AI · 2026-11-04
- YANAMI: 400K-dialog multimodal dataset shows models still fail at stateful multiturn interaction — Shengyu Wang · 2026-11-04
- Open-source Idol model turns a single image into 3D in 45s on 8GB VRAM — Promptmethus · 2026-11-04
- ComfyUI v0.44.0 released with new Patina texture nodes — Fcking_Chuck · 2026-11-04
- Krea 3 technical report released: a deep dive into data, architecture and training — bdsqlsz · 2026-11-04