Can't the model be faster than 30 min to generate a 3 min. song?
I know it's not your model and you did your best to bring it to ComfyUI.
However, MiniMax Music model requires 50 step to generate a song while ACE-STEP turbo only requires 8 steps and it generates a 3 min song in a minute and half on my PC with highest quality that even beats MinMax. Seriously why do you even consider adding MiniMax Music model? Which is again a model that is not optimized for consumer grade GPUs.
Well, once you handle MM3 the right way it produces audio that is superior in many ways compared to ACE1.5. But handling is more complicated and, yes, more time consuming.
Try this ComfyUI custom node:
https://huggingface.co/Comfy-Org/MiniMax-Music-3/discussions/16
It produces satisfying and surprising superior and clear results pretty much every time you use it. And it works on my consumer level PC: 9400 CPU, 64GB RAM, 5060 ti 16GB VRAM. One 4-5 min song take around 8-12 minutes, but it's worth the wait!