YingMusic-Singer-Plus-ComfyUI

Voice/melody-conditioned singing generation and lyric replacement, using native ComfyUI components.

This is a converted model package for Jack Spooner’s ComfyUI fork, derived from ASLP-lab/YingMusic-Singer-Plus. It is not a new fine-tune.

Required code

The integration modifies ComfyUI core; this is not a drop-in custom_nodes package.

• Model integration commit
• Implementation · Offline converter
• Development branch · Pinned compatible reference snapshot 9d9053c5e
• Installation, node usage and acceptance details

For an existing ComfyUI installation, use the linked core changes and their prerequisites against a compatible base. The integration commit alone is not asserted to be an independently cherry-pickable patch. The pinned reference snapshot includes the required shared loaders, attention and memory changes. Follow the fork’s installation README and dependencies when reproducing that snapshot; update the application only while its queue is idle. Upstream ComfyUI is Comfy-Org/ComfyUI; its README describes the underlying node engine and model management.

Files and installation

Copy the model category folders into ComfyUI/models/, preserving their names and subfolders. Supporting pronunciation/configuration assets in this repository are part of the package.

File	Size
audio_encoders/yingmusic_singer_plus.safetensors	0.471 GB
diffusion_models/yingmusic_singer_plus.safetensors	1.344 GB
text_encoders/yingmusic_singer_plus/g2p_chinese_model/poly_bert_model.onnx	0.077 GB
vae/yingmusic_singer_plus.safetensors	0.624 GB

Manifest/provenance files record the conversion and source revisions. SHA256SUMS and FILES.json describe this release. Filenames and model folders matter. No original source snapshot needs to be converted again.

Usage

Copy the diffusion_models, audio_encoders, vae and text_encoders folders into the corresponding ComfyUI model folders. Use standard diffusion/audio-encoder/VAE loaders, then YingMusic Generate / Edit Singing. Supply isolated voice and melody AUDIO references, their matching reference lyrics, and target lyrics. Baseline: 32 Euler steps, guidance 3, time shift 0.5 and reference silence 0.5 s. Install the pinned fork’s requirements-yingmusic.txt and system espeak-ng as documented. The optional Chinese ONNX G2P assets are included in text_encoders/yingmusic_singer_plus.

Verification and limits

Bounded native CPU/GPU generation, cancellation/reload and RouterAudio reference editing are documented. This is not a multilingual singing-quality or universal VRAM benchmark. Model duration and memory depend on the supplied references. Reference lyrics must describe the reference audio. Separation and accompaniment mixing are separate operations.

These statements describe the linked retained tests, not a new cross-platform acceptance suite. File size is not a minimum RAM/VRAM requirement. No inference-time downloads or external inference worker are required by these native model paths.

Conversion and provenance

Splits the original non-distilled EMA checkpoint into singing transformer, melody encoder and Stable Audio VAE; excludes unused student MIDI weights. Preserves original FP32 component tensors and copies vocabulary/pronunciation assets from pinned source code.

Source model revision: 9b3f444f2bccd77fbc03e32eb7c334fc96040b8d. See PROVENANCE.json, the conversion manifest and NOTICE. No training data, private inputs or generated user recordings are included.

License and attribution

Main YingMusic weights are CC BY 4.0, with attribution and modification notice. The included VAE is separately covered by Stability AI Community License (5 July 2024); it is NOT relicensed as CC BY. Powered by Stability AI. Commercial use/distribution of Stability materials has registration/revenue conditions; consult LICENSE-STABILITY and the Stability AI AUP. Amphion G2P credits/notices are retained.

See LICENSE, NOTICE, and the retained upstream model card/source notices. Model authors and publishers retain credit; this is a third-party ComfyUI repackaging and does not imply endorsement. The weights’ terms and the separately linked ComfyUI code license are distinct.