Free ComfyUI templates chain local LLM lyrics into YuE2 song rendering, runs on 12GB VRAM
Vivid_Promise1700 · reddit · 2026-10-20
A developer released a free set of ComfyUI music production templates (Plenio Music Production System, Apache 2.0) built around ComfyUI's native YuE2 and MiniMax Music 3 nodes for fully local song generation.
The YuE2 Song graph:
- A Song Brief feeds a local LLM writer (Gemma 4, or any GGUF / LM Studio / Ollama model) that drafts lyrics
- A Song Sheet node shows the exact style, lyrics and ABC score YuE2 receives; all three are editable with an optional approval stop
- YuE2 plans and renders, then EQ, loudness and tagged FLAC/MP3 export; the workflow is embedded in the FLAC — drop the file back into ComfyUI to continue the song
- New in 1.0.2: a creativity slider where 50 equals native defaults, moderating writer rules and YuE2 sampling (temperature, top-p, top-k)
Also included: covers of recordings, MiniMax Music 3, hand-writing melodies in a score editor, and a mastering-only graph. Requires ComfyUI 0.37+ (tested on 0.41), no extra Python packages. The int8 YuE2 model runs well on 12-16GB VRAM; keep songs short on 8GB. Note YuE2 weights are CC BY-NC (non-commercial).
More from Multimodal
- 59 animations in a 43-min video: creator breaks down his AI agent workflow — MaxLenormand · 2026-10-20
- Dev generates an entire desert city with AI, says it would make great FPS maps — jasonkneen · 2026-10-20
- Feeding the whole Animator's Survival Kit to Opus 5.5 yields decent keyframe animation — MoonL88537 · 2026-10-20
- Glimpse launches as a visual thinking model claiming Astra-level reasoning benchmarks — LinusEkenstam · 2026-10-20
- ComfyUI lets you pipe images into an LLM — how do you do the same with audio? — PhilosopherSweaty826 · 2026-10-20
- AI-generated mecha battle video looks straight out of a movie — Key_Layer6250 · 2026-10-20