Model Card for VRoid Diffusion

<!-- Provide a quick summary of what the model is/does. -->

This is a latent text-to-image diffusion model to demonstrate how U-Net training affects the generated images.

• Text Encoder is from OpenCLIP ViT-H/14, MIT License, Training Data : LAION-2B
• VAE is from Mitsua Diffusion One, Mitsua Open RAIL-M License, Training Data: Public Domain/CC0 + Licensed
• U-Net is trained from scratch using full version of VRoid Image Dataset Lite with some modifications.
• VRoid is a trademark or registered trademark of Pixiv inc. in Japan and other regions.

Model Details

• vroid_diffusion_test.safetensors
  • base variant.
• vroid_diffusion_test_invert_red_blue.safetensors
  • red and blue in the caption is swapped.
  • pink and skyblue in the caption is swapped.
• vroid_diffusion_test_monochrome.safetensors
  • all training images are converted to grayscale.

Model Variant

• VRoid Diffusion Unconditional
  • This is unconditional image generator without CLIP.

Model Description

• Developed by: Abstract Engine.
• License: Mitsua Open RAIL-M License.

Uses

Direct Use

Text-to-Image generation for research and educational purposes.

Out-of-Scope Use

Any deployed use case of the model.

Training Details

• Trained resolution : 256x256
• Batch Size : 48
• Steps : 45k
• LR : 1e-5 with warmup 1000 steps

Training Data

We use full version of VRoid Image Dataset Lite with some modifications.