Super-Resolution with Perturbed-Attention Guidance

This repository is based on Diffusers.

The pipeline is a modification of StableDiffusionPipeline to support super-resolution with Perturbed-Attention Guidance (PAG). Please refer to "Image-to-upscaler-to-super-resolution" section of an official document for details.

Quickstart

Loading Custom Piepline:

from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-x4-upscaler",
    custom_pipeline="hyoungwoncho/sd_perturbed_attention_guidance_sr",
    torch_dtype=torch.float16,
    safety_checker=None
)

device="cuda"
pipe = pipe.to(device)

Super-Resolution with PAG:

output = pipe(
    prompts,
    image=lr_image,
    num_inference_steps=50,
    guidance_scale=0.0,
    pag_scale=2.0,
    pag_applied_layers_index=['u2']
).images[0]

Parameters

guidance_scale : gudiance scale of CFG (ex: 7.5)

pag_scale : gudiance scale of PAG (ex: 2.0)

pag_applied_layers_index : index of the layer to apply perturbation (ex: ['u2'])