Hey Jerome. Do you have at least 16GB of VRAM? The models may work with 12GB but inference would likely be painfully slow so I decided to make them available only to larger cards. I should have made this clearer in the post so my apologies.
Having said that, I could make the models available on 12GB cards if you're keen on trying them out and are willing to put up with potential slowness.
The idea is to use existing LoRa modules because, on smaller systems, this would allow us to add a specific LoRa module to the basic model, depending on the style of music we want to generate. For example, we should normally be able to use this: https://civitai.com/models/2425543/ace-step-15-music-generation-for-low-vram or https://civitai.com/models/2624128/ace-step-15-psytrance-lora-ver-20... Indeed, there's an option to create your own LoRA modules in an hour with a simple setup, as described here: https://github.com/ace-step/ACE-Step-1.5/blob/main/docs/en/LoRA_Training_Tutorial.md This is definitely an option that could be of interest to other users of your application.
Indeed, the first link is incorrect; I confused it with this one: https://civitai.com/models/2642635/minimal-techno-overdrive-lora-ace-step-v15-xl-base But it requires a 12GB card.
