pearisli/ldm-lsun-church

Latent Diffusion Models (LDM)

Model Details

Model Description

The model was resumed for 150,000 steps starting from the LDM-KL-8 checkpoint to incorporate the attention mechanism as implemented in diffusers, which differs from the original version. It leverages the autoencoder from stable-diffusion-2-base and achieves the FID-50K score of 4.63.

Note: To use this model with diffusers, replace the file diffusers/models/unets/unet_2d.py with the unet_2d.py provided in this repository.

Uses

Inference with a pipeline

from diffusers import LDMPipeline

model_id = "pearisli/ldm-lsun-church"

# load model and scheduler
pipeline = LDMPipeline.from_pretrained(model_id)

# run pipeline in inference (sample random noise and denoise)
image = pipeline(num_inference_steps=200).images[0]

# save image
image.save("ldm_generated_image.png")