Krea2 LoRA
This repository contains an open-source LoRA model trained mostly for Japanese characters and compatible with Krea2.
Viewer: https://huggingface.co/spaces/fidel1234/Krea2_Asian_Character_Showcase
📝 Model Details
License: MIT License (Note: Users must also comply with the base Krea 2 Community License Agreement and Acceptable Use Policy when using the underlying Krea base models).
Base Model: Krea2
Type: LoRA (Low-Rank Adaptation)
Training Framework: Trained on Krea2 raw using AI-Toolkit
Compatibility: Designed to work smoothly on both ComfyUI raw and turbo checkpoints.
Dataset Quality & Framing: Trained on a hand-cherry-picked dataset featuring a balanced mix of close-up, medium shot, full-body shot, and different angles for flexible framing and perspective control.
🖼️ Assets & Workflow
Sample Image: Each LoRA file is accompanied by a sample preview image matching the exact filename of the .safetensors model (e.g., model_name.safetensors and model_name.png).
Embedded Workflow: The generation workflow and trigger word are embedded only inside the metadata of the sample image (not the safetensors file). You can drag and drop the sample image into your UI (such as ComfyUI) to load the complete workflow, parameters, and trigger word automatically.
🚀 Usage & Recommendations
Inference: Load the model and drag the provided sample image into your workflow tool to inspect or use the exact generation parameters.
Recommended Weight: 0.8 - 1.0
Inference Steps: 20 - 30
🛡️ Terms of Use & Safety
This model is released under the MIT License.
Permitted Use: You are free to use, modify, share, and commercialize this model according to the terms of the MIT License.
Disclaimer: Please use them responsibly and respect the rights of the individuals depicted - don't create misleading, harmful, or non-consensual content. By downloading, you take responsibility for what you generate. Use as is without harmful intent.