all workflows not labeled 1.0 are up to date and working. give em a try.
acestep 1.5 nodes optimized for recent version of comfyui, ping me with issues I will usually fix in a few minutes if im awake.
TUTORIALS UNDER VERSION----->
Hey everyone,
Wanted to share some nodes I've been working on that unlock the full ACE-Step 1.5 feature set in ComfyUI.
What's different from native ComfyUI support?
ComfyUI's built-in ACE-Step nodes give you text2music generation, which is great for creating tracks from scratch. But ACE-Step 1.5 actually supports a bunch of other task types that weren't exposed - so I built custom guiders for them:
- Edit (Extend/Repaint) - Add new audio before or after existing tracks, or regenerate specific time regions while keeping the rest intact
- Cover - Style transfer that preserves the semantic structure (rhythm, melody) while generating new audio with different characteristics
- (wip) Extract - Pull out specific stems like vocals, drums, bass, guitar, etc.
- (wip) Lego - Generate a specific instrument track that fits with existing audio
Time permitting, and based on the level of interest from the community, I will finish the Extract and Lego task custom Guiders. I will be back with semantic hint blending and some other stuff for Edit and Cover.
-
Part of [ComfyUI_RyanOnTheInside](https://github.com/ryanontheinside/ComfyUI_RyanOnTheInside) - install/update via ComfyUI Manager.
Let me know if you run into any issues or have questions and I will try to answer!
Love
Ryan
Description
Blend semantic hints from two input songs. tutorial:
FAQ
Comments (11)
AudioInfo
Numba needs NumPy 2.3 or less. Got NumPy 2.4.
You can either upgrade numba or downgrade numpy
thank you for reporting. Turned out to be a problem for a bunch of people; i totally removed librosa from the pack so this is a non issue now
Your nodes are awesome, thanks for the crazy fast update. Ace-step still feels unwelcome on this site. Hard to search, uncitable as a base model, hard to tag, etc. Annoying. Anyway, I could never get the Acestepgen-style nodes to recognize anything- I've probably got ten copies of all the models in every folder and subfolder by now. So your native-compatible nodes are just the bee's knees. I don't know why it's like pulling teeth to get people interested in musical slop, they clearly love every other kind of slop. Whatever, thanks for sharing your work. My only request is that you make your console logo bigger. It's a little hard to read at the moment.
I don't know how you're doing it, but this is really, really fast. And repainting is so much more stable with your nodes, adherent to prompting, staying in key, everything. Really cuts down on the tedious iteration. You nailed that so fast. Friggin awesome.
THANK YOU!!! <3 You are kind
Big update tomorrow, still no Lego or Extract task working, but its sick
Update did break my native nodes, but I see this has come up on the git page so I'm not going to whine about something that I'm sure you are going address. Just a heads-up. Had to roll back to restore base functionality and set up the current version on a separate install until it's fixed.
Very nice nodes, I have one question, is it posible to generate vocals for my music that has no vocals in it ? how to call this process in english ?:) its not cover neither lego neither blend.. :D
I forgot to mention that i have the lyrics..
@poisas69220 I would try "[acapella]" (I haven't tried it yet), and then mix it in a DAW
I'm struggling to make vocals appear in parts that are just instrumental, any hint for that in repaint option?