ComfyUI-Olm-YuE2
A set of ComfyUI custom nodes built on the YuE2 music generation model. The generation pipeline is divided into four intervenable stages—Plan, Semantic, Synthesize, and Decode—and supports ABC notation score editing, providing a stage-by-stage controllable AI music creation workflow.
Timeline (last 90 days)
ComfyUI-Olm-YuE2 is a custom node set built by developer o-l-l-i that integrates the YuE2 music generation model into ComfyUI workflows.
ComfyUI-Olm-YuE2 can generate 48 kHz stereo songs by simply providing a music style description and lyrics.
ComfyUI-Olm-YuE2 splits the music generation process into four stages—Plan, Semantic, Synthesize, and Decode—where the intermediate results of each stage can be inspected, saved, reused, and branched.
The project provides an optional score editing UI that includes staff notation rendering, an ABC score view, and an ABC editor node, which needs to be enabled in ComfyUI settings.
The score UI is entirely optional; regular node execution and API workflows do not depend on it.
ComfyUI-Olm-YuE2 reuses the Torch/Transformers bundled with ComfyUI and only requires the additional installation of two packages: tiktoken and accelerate.
The project's test environment is CUDA v13.0 and Python 3.13, and the installation process has been verified on a fresh installation of ComfyUI.
The author has so far only conducted full testing on a single RTX 5090, and measured VRAM usage data is provided in the README.
The project has a built-in experimental VRAM offloading mechanism (demonstrated in workflow 06) that can significantly reduce memory usage, but memory spikes still occur during the process.
The author is researching optional quantization schemes to further save VRAM, but the trade-off between memory savings from quantization and audio quality still needs to be tested, especially in the synthesis stage.
2 more timeline events
All Facts (10)
ComfyUI-Olm-YuE2 is a custom node set built by developer o-l-l-i that integrates the YuE2 music generation model into ComfyUI workflows.
50%UnverifiedComfyUI-Olm-YuE2 can generate 48 kHz stereo songs by simply providing a music style description and lyrics.
50%UnverifiedComfyUI-Olm-YuE2 splits the music generation process into four stages—Plan, Semantic, Synthesize, and Decode—where the intermediate results of each stage can be inspected, saved, reused, and branched.
50%UnverifiedThe project provides an optional score editing UI that includes staff notation rendering, an ABC score view, and an ABC editor node, which needs to be enabled in ComfyUI settings.
50%UnverifiedThe score UI is entirely optional; regular node execution and API workflows do not depend on it.
50%UnverifiedComfyUI-Olm-YuE2 reuses the Torch/Transformers bundled with ComfyUI and only requires the additional installation of two packages: tiktoken and accelerate.
50%UnverifiedThe project's test environment is CUDA v13.0 and Python 3.13, and the installation process has been verified on a fresh installation of ComfyUI.
50%UnverifiedThe author has so far only conducted full testing on a single RTX 5090, and measured VRAM usage data is provided in the README.
50%UnverifiedThe project has a built-in experimental VRAM offloading mechanism (demonstrated in workflow 06) that can significantly reduce memory usage, but memory spikes still occur during the process.
50%UnverifiedThe project does not include model weights; users need to download the YuE2-related files themselves and place them in the designated location.
50%