anime-faces-ddpm

A Denoising Diffusion Probabilistic Model (DDPM) trained to generate anime faces using this example as a basis.

Model Description

This diffusion model is trained with the 🤗 Diffusers library on the huggan/anime-faces dataset.

How To Use

from diffusers import DDPMPipeline

checkpoint = "sweetfelinity/anime-faces-ddpm"
pipeline = DDPMPipeline.from_pretrained(checkpoint)
pipeline = pipeline.to("cuda") # or "cpu"

for i in range(10):
    image = pipeline().images[0]
    image.save(str(i + 1) + ".png")

Training Hyperparameters

The following hyperparameters were used during training:

• resolution=64
• train_batch_size=16
• num_epochs=30
• gradient_accumulation_steps=1
• learning_rate=1e-4
• lr_warmup_steps=500
• mixed_precision=fp16
• checkpointing_steps=2000
• save_images_epochs=4
• use_ema
• adam_weight_decay=1e-6
• lr_scheduler=linear
• eval_batch_size=32

Training Results

See model folder /generated-images for 100 images created by the DDPM.