Got it going on my win10 box. Had to downgrade to the 0.1.10 version, 0.1.12 didn't wanna work for me. Their cuda support is buggy, but 45 seconds of audio took about 10 minutes on my CPU. Not sold on quality of the qwen model yet. But to be fair I used kind of a jokey audio. It generates the speech fine but I think my jokey sample may have confused its ability to handle tonality. I'll give it a couple more shots. I see potential, at least. Those screenshots showing timeline and track editing aren't real tho and I question their judgement on sharing that.