ACE-Step 1.5 LoRA: German Folk Metal

This is a LoRA fine-tune for the ACE-Step 1.5 model, specifically trained to generate German Folk Metal. The model captures the high-energy fusion of aggressive metal instrumentation (distorted guitars, double-bass drums) and traditional folk elements (hurdy-gurdy, bagpipes) with characteristic German-language vocal delivery.

Model Details

Model Description

This LoRA was trained to adapt the ACE-Step 1.5 base model to the specific aesthetic, production style, and instrumentation of the German Folk Metal genre. It is optimized to generate tracks with high dynamic range, tavern-like atmosphere, and rhythmic folk-metal intensity.

  • Model type: ACE-Step 1.5 LoRA (Low-Rank Adaptation)
  • Language(s): German (primarily), Multi-lingual capability preserved
  • Finetuned from model: ace-step-1.5-turbo

Uses

Direct Use

The model is intended to generate high-quality folk-metal music compositions. It is best used for tracks requiring a mix of heavy metal elements and traditional acoustic folk instrumentation.

Trigger Keyword

To activate the style, ensure your prompt begins with the trigger keyword: german_folkmetal

How to Get Started with the Model

Use the following style guide for optimal generation:

Recommended Prompt Structure:

german_folkmetal, Folk Metal, [high-energy], [distorted electric guitars], [traditional hurdy-gurdy melody], [driving double-kick drums], [powerful male vocals], [bagpipes]

Training Details

Training Data

The model was trained on a high-quality, curated dataset of 25 tracks from established German Folk Metal artists (e.g., Feuerschwanz, Saltatio Mortis, Harpyie, In Extremo). The dataset was preprocessed for consistency, normalizing genre tags to "Folk Metal" to ensure stable latent mapping.

Training Procedure

Training Hyperparameters

  • Learning Rate: 0.00001 (1e-5)
  • Batch Size: 3
  • Gradient Accumulation: 3
  • Target Epochs: 500
  • Training regime: bf16 mixed precision

Speeds, Sizes, Times

  • Hardware: Consumer GPU (RTX series)
  • Training Time: ~1.5 hours total (estimated)

Bias, Risks, and Limitations

  • Limitations: The model is optimized for the German Folk Metal genre; generation of other genres may lead to style bleeding or suboptimal audio fidelity.
  • Biases: The model inherits the linguistic and cultural biases of the German Folk Metal scene (e.g., specific lyrical themes involving tavern culture, mythology, and folklore).

Technical Specifications

Model Architecture and Objective

The LoRA applies low-rank adaptations to the Diffusion Transformer (DiT) weights of the ACE-Step 1.5 architecture, allowing for precise style transfer without requiring full-parameter fine-tuning of the frozen Language Model planner.


Model Card generated for German Folk Metal LoRA.

Downloads last month
80
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support