It always gives random stuff that does not sound good. Using same prompt structure from huggingface space and caption with the LLM turned off and it gives nothing like with the Ace-Step Music Generator Node and api key. Tried with sample_mode on/off and use_format on/off all just random stuff that has nothing to do with the prompting is returned.