Summary
This tutorial from AI Search covers UA2, a newly released open-source AI music generator that runs on as little as 4 GB of VRAM. The video walks through what makes UA2 stand out: the ability to generate songs from text prompts, create cover songs by feeding in a reference audio track, edit existing songs (changing key, style, or instrumentation), and do all of this locally without relying on a cloud service.
The demonstration includes genre examples spanning smoky jazz, 1930s boogie woogie, Spanish flamenco, Russian folktronica, and Korean emo, showing multilingual lyric support across the board. A particularly notable feature is UA2’s internal score-based representation — rather than going directly from prompt to audio, it first writes out an editable musical score (notes, key, tempo) before rendering the final track. This architecture is what enables its flexible cover and editing capabilities. The tutorial also shows UA2 integrated with an AI agent (GPT/GLM) for iterative music editing via text prompts.
On benchmarks, UA2 reportedly outperforms other open-source models like Minimax Music and a competing model, and edges out Suno V6 on a song quality index. The full installation walkthrough uses ComfyUI, covering model downloads (a 7.8 GB BF-16 checkpoint or a 3.96 GB quantized alternative), workflow setup, and both the text-to-music and audio-reference pipelines. A useful watch for anyone exploring local AI media generation.
📺 Source: AI Search · Published September 15, 2026
🏷️ Format: Tutorial Demo
