Behold, workflow whisperers!
Generate music with the powerful and lightweight YuE2 model using this easy-to-use ComfyUI workflow.
Features
App mode enabled by default: a friendly UI that helps you focus on the most important parameters only.
Native nodes only: no need to install custom nodes.
Models auto-download: this workflow is configured to make ComfyUI prompt you to download missing models from their original sources.
Speed boost included: KSampler configured to run 16x faster than default templates while keeping quality.
Enhance with AI: just toggle a button and you can have your style and lyrics automatically enhanced by a local and lightweight LLM (Gemma 4 e2b by default).
Reimagine existing songs: provide a reference audio to generate a cover - listen to your favourite song in a different style or with different lyrics.
Style/genre presets: not sure which genre you're looking for? Pick one from a list of 50+ popular styles.
Quality LoRAs included: knobs to control influence of the quality LoRAs that improve your results.
Edit song tempo: adjust whether you want the same song to be faster or slower.
Feel free to provide constructive criticism and feedback so we can all learn.
Until the next Supernova! [End transmission.]
Description
v1.0 - Drag the workflow json file into ComfyUI to import it. You should see the app form instead of the node graph (otherwise, press Alt + M to toggle app mode).
FAQ
Comments (9)
Hello. Do you have a non-UI non-subgraphed version ? pleaaase? 🪵🐸🪵
Do you mean a script that you can run headless? Or a ComfyUI workflow with no App mode and no subgraph? For the latter, you can just use Comfy's YuE2 templates and unpack the subgraph.
@SupernovaTech Yes I meant an "exploded" ComfyUI workflow so I can try to understand how it works. "For the latter, you can just use Comfy's YuE2 templates and unpack the subgraph." Where can I find that?! 🙌
@FrogOnStilts Got it. Once you update your ComfyUI, press "Templates" on the left sidebar, search for "YuE2" then select one of the templates. Both come with a subgraph, you just need to right click it and select "Unpack subgraph".
Hi, when doing covers, is there a way to make the source be less influential? So it becomes less like the source and more creative?
Also "Tempo override" in the Edit score group, how is that to use?
One of the form fields at the end called "value" (a bug in ComfyUI, it should be called "Generation mode") is set to "full" by default, meaning it will use the full song as a reference. Changing it to "melody" will keep the melody/vocal, and generate something different for the harmony (chords/base).
I'm afraid that's all we can do for now. YuE2 doesn't have released an official encoder, meaning we can't control % of influence (similar to how we control image denoise). Hopefully the community will figure out a way to do that though.
Regarding tempo override, you can change it to something like < 80 to make the song very slow, or > 140 to make it very fast, or leave the default value (0 = auto) to let the model choose it based on your prompts.
@SupernovaTech Thank you. I'll give it a try.
It doesn't find the Style preset prefix.
And how do I disable the Cover mode? It just needs generation.
Not sure what you mean, I just imported the workflow in a brand new ComfyUI instance and the style presets is working as intended. Also, cover mode is disabled by default, you only enable it if you set > 1s duration in your reference song. Did you download the workflow from the correct source (Download button)?