UPDATE:
I managed to merge in a couple of different LORAs which made a huge difference in it's "abilities".
I also released the python script to extract your own Audio weights.
After some experimentation I managed to extract the Audio DiT Layers of Sulfur and get them working in Dramabox.
The result is some added Audio generation "features" - but prompting is a bit tricky and I am still figuring it out :). This is VERY much a WIP.
Usage:
This is a drop-in replacement for the dramabox-dit-v1.safetensors file.
When using Huggingface for Dramabox, first find the Cache where the model is stored.
Once you found that, go to the "blobs" subfolder.
There will be one 6.6GB file with a hash as a name.
Replace that file with this and make sure to give it the EXACTLY same name.
You should now be ready to go!
Alternatively you can of course just specify the path to this file instead of the original DiT in your code :)
I am still experimenting with the audio components part as they seem to have modified some things there.
Not sure if they even make a difference if I manage to change them aswell.
