metadata
language:
- en
pipeline_tag: unconditional-image-generation
tags:
- Diffusion Models
- Stable Diffusion
- Perturbed-Attention Guidance
- PAG
Perturbed-Attention Guidance
This repository is based on Diffusers. StableDiffusionPAGPipeline is a modification of StableDiffusionPipeline to support Perturbed-Attention Guidance (PAG).
Quickstart
Load StableDiffusionPAGPipeline as below:
pipe = StableDiffusionPipeline.from_pretrained(
"runwayml/stable-diffusion-v1-5",
custom_pipeline="hyoungwoncho/sd_perturbed_attention_guidance",
torch_dtype=torch.float16,
safety_checker=None
)
Sampling:
output_baseline = pipe(
prompts,
width=512,
height=512,
num_inference_steps=50,
guidance_scale=0.0,
pag_scale=4.5,
pag_applied_layers_index=pag_applied_layers_index
).images
Parameters:
pag_scale : gudiance scale of PAG (ex: 4.5) pag_applied_layers_index = index of the layer to apply perturbation (ex: ['m0'])