| Property | Value |
|---|---|
| License | Fair AI Public License 1.0-SD |
| Training Dataset | 7,467 images from p1atdev/pvc |
| Base Model | Waifu Diffusion v1.5 beta 2 |
| Available Formats | FP16/FP32 Safetensors/CKPT |
What is pvc-v3?
PVC-v3 is a specialized latent diffusion model designed for generating high-quality anime-style PVC figure images. It's built on top of Waifu Diffusion v1.5 beta 2 and supports various rendering styles including standard PVC, figma, and nendoroid aesthetics.
Implementation Details
The model was trained using the Kohya trainer on an A5000 GPU, with 5 epochs and a resolution of 896. It implements the Lion optimizer with a learning rate of 4e-7 and uses a cosine_with_restarts scheduler.
- Supports multiple file formats including FP16 and FP32 variations
- Requires WD's VAE for optimal results
- Implements xformers memory-efficient attention
Core Capabilities
- Generation of PVC figure-style images with anime aesthetics
- Support for Danbooru tags in prompts
- Multiple style options: standard PVC, figma (with joints), and nendoroid (chibi style)
- High-quality output with customizable parameters
Frequently Asked Questions
Q: What makes this model unique?
The model specializes in generating PVC figure-style images with precise control over aesthetics through different trigger words (pvc, figma, nendoroid) and supports detailed customization using Danbooru tags.
Q: What are the recommended use cases?
The model is ideal for generating anime-style figure illustrations, product-like figure renders, and chibi-style character artwork. It's particularly effective when combined with specific style triggers and detailed prompting.