ACE-Step-v1.5-ambient_dream1-LoRA

Model Card for Model ID:

LoRA adapter with ambient sounds for ACE-Step/Ace-Step1.5

Model Description:

In file "start_gradio_ui.bat" I used these:<br> set CONFIG_PATH=--config_path acestep-v15-turbo<br> set LM_MODEL_PATH=--lm_model_path acestep-5Hz-lm-4B<br> set INIT_SERVICE=--init_service false<br>

init_service false, so I could get access to the LoRA loading part in the Web UI.<br> (The models are also loaded with Web UI when this is set)<br>

The LoRA adapter is for ambient sound for the acestep-v15-turbo song generator ai model.<br> It works best with pure ambient songs, not so good with pop, rock etc songs.<br>

Trigger word in caption is:<br> ambient_dream1<br>

See the demo songs and the .json files. The .json files include all settings and full caption used for each song.<br> Again, LM model was acestep-5Hz-lm-4B together with DiT model acestep-v15-turbo, both used for LoRA training and song generation.<br>

Model Sources:

https://huggingface.co/ACE-Step/Ace-Step1.5<br>

Uses:

LoRA adapter for ACE-Step/Ace-Step1.5, ambient sounds, instrumental<br>

Training Details:

Trained on 10 short instrumental songs made and recorded by myself.<br> In total about 12 minutes of audio. Ambient sounds.<br> Trained 600 epochs. Took about 14 houres on a Nvidia RTX 3080 Ti laptop, 16 GB VRAM.<br>