PVC v2
| Property | Value |
|---|---|
| License | CreativeML OpenRAIL-M |
| Base Model | Waifu Diffusion v1.4 epoch 2 |
| Training Images | 2,110 |
| Resolution | 768x768 |
What is PVC?
PVC is a specialized text-to-image diffusion model fine-tuned specifically for generating high-quality anime-style character illustrations. Built upon Waifu Diffusion v1.4, it has been trained on a curated dataset of PVC figure images using the LoRA (Low-Rank Adaptation) method, enabling it to produce detailed character renders with impressive consistency.
Implementation Details
The model was trained using the kohya_ss/sd-scripts framework with carefully optimized parameters including a U-Net learning rate of 1e-4 and text encoder learning rate of 5e-5. It utilizes mixed precision training (fp16) with xformers optimization enabled.
- Training Steps: 21,100
- Network Dimension: 16
- Optimizer: AdamW8bit
- Max Token Length: 225
Core Capabilities
- High-quality anime character generation
- Support for Danbooru-style tag prompting
- Specialized in detailed character illustrations
- Optimized for 768x768 resolution outputs
Frequently Asked Questions
Q: What makes this model unique?
PVC v2 specializes in generating high-quality anime character illustrations with a particular focus on figure-like aesthetics, thanks to its specialized training dataset of PVC figure images.
Q: What are the recommended use cases?
The model excels at generating detailed character illustrations, particularly when using quality tags like "masterpiece, best quality" at the start of prompts. It's ideal for creating anime-style character concepts and illustrations.
