Suno and Udio may finally have some open source competition. Here I take a quick look at using the new M-A-P YuE music generation model via the YuE-Extend project. This is a lyrics-to-mixed audio model that can take in annotated lyrics and output full-length, mixed songs.
This supports exllamav2 quantized models and music extension (use the -icl models. I don't cover extension in the video, but the GUI has instructions). This is mostly just an introduction to this handy project by Mozer with plenty of examples of the YuE model being used throughout the video. All music used was generated by M-A-P YuE (mostly the exl2-8 model).
I forgot to note it in the video - if you generate a track you like in the -extend WebUI, save the stems (vocal/instrumental) from the output folder before they are mixed. Theyre overwritten on every generation. The final tracks have timestamped names, but the intermediate tracks will be lost if you dont copy them.
0