🎧 Smoki Lofi - ACE-Step 1.5 LoRA
This is a specialized Low-Rank Adaptation (LoRA) for the ACE-Step 1.5 music generation model. It was trained to capture a specific "warm and dusty" Lo-Fi aesthetic, moving the base model away from clinical digital sounds toward a more organic, sampled vibe.
🌟 Key Features
- Enhanced Texture: Injects vinyl crackle, tape saturation, and mid-range warmth.
- Instrument Character: Softens piano transients and adds "weight" to hollow-body jazz guitars.
- Cohesive Mix: Improves the "sonic glue" between instruments for a more professional sampled feel.
🚀 Usage Instructions
Trigger Words
Use the following tag in your prompt to activate the style:
- Primary Tag:
smoki_lofi
- Recommended Supporting Tags:
mellow piano, jazz guitar, nostalgic atmosphere, dusty beats, [Instrumental]
Recommended Settings
- LoRA Strength:
0.7 - 0.9 for a subtle flavor.1.5 - 2.0 for the full Pixabay-inspired character (recommended in ComfyUI).
- Sampling:
- AutoFlow Shift:
3.0
- Steps:
12
🧩 ComfyUI Integration
- Download
adapter_model.safetensors.
- Place it in your
ComfyUI/models/loras folder.
- Use the Load LoRA node.
🛠️ Training Metadata
- Dataset: 43 High-quality curated Lo-Fi tracks (Pixabay sourced).
- Hardware: NVIDIA GeForce RTX 5090 (24GB+ VRAM utilized).
- Learning Rate:
0.00005 (5e-5) for high detail retention.
- Epochs: 100
- Loss Convergence: ~0.13 (High fidelity).
⚖️ License
This model is released under the Apache 2.0 license. You are free to use the outputs for commercial or non-commercial projects, but the model itself is provided "as-is".
