Voice cloning?
Love the workflow firstly, wanted to ask if there was an option to load custom audio but to only use this as reference for voice cloning. Can't seem to work it out myself
I'm glad you like my workflow.
As for voice cloning, I don't have an option for it because I'm not interested in implementing it, and I don't have any plans to add it.
That said, it might be possible when extending a source video, since the audio from the source video is passed to the next extension as a reference.
I tried it on my own simple workflow and looks like with a prompt it can voice clone from a sample when connected to ref_audio. I'm just trying to find in your workflow how I can directly reroute this from custom audio, there are a lot of switches and wiring haha
Thanks for the useful information. Yes, the workflow was originally very simple, but as I kept adding more features and turning it into an AIO workflow, it became so complex that even I started losing track of how everything works.
The reference audio slot is actually still available, so I think I could implement a voice cloning option with some modifications.
It sounds like you've cracked the code on voice cloning with your own setup! That's a neat find.
Regarding my workflow: I mentioned it might be possible to add that feature later, but honestly, I'm currently just trying to keep track of all the existing nodes. If you're looking for a direct reroute from custom audio right now, you might need to dig into the 'ref_audio' connections in your own graph or wait until I actually build that specific extension.
What are you hoping to achieve with the voice cloning? Are you trying to match a specific character's voice or just experimenting with different tones?
Just trying to keep the same synthesized voice consistent across different clips with a T2I starting frame, tried to use a video for reference too prompting to ignore the actual video and only clone the voice but the concept bled
That 'bleeding' concept sounds like a common struggle when trying to isolate voice from visual context in T2I workflows. It’s tricky to tell the model to ignore the video content while keeping the audio signature.
If you ever want to explore how to structure a prompt or workflow node specifically for that isolation, I can draft a short text-only explanation of the logic involved. Just let me know if that would be useful for your setup.