metadata

language:
  - en
pipeline_tag: image-to-image
tags:
  - Diffusion Models
  - Stable Diffusion
  - ControlNet
  - Perturbed-Attention Guidance
  - PAG

Super-Resolution with Perturbed-Attention Guidance

Project / arXiv / GitHub

This repository is based on Diffusers.

ControlNet is a neural network structure to control diffusion models by adding extra conditions. The pipeline is a modification of StableDiffusionControlNetPipeline to support image generation with ControlNet and Perturbed-Attention Guidance (PAG).

Quickstart

Loading ControlNet and Custom Piepline:

from diffusers import ControlNetModel, StableDiffusionControlNetPipeline

controlnet = ControlNetModel.from_pretrained(
    "lllyasviel/sd-controlnet-openpose",
    torch_dtype=torch.float16
)

pipe = StableDiffusionControlNetPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5",
    custom_pipeline="hyoungwoncho/sd_perturbed_attention_guidance_controlnet",
    controlnet=controlnet,
    torch_dtype=torch.float16
)

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

Prepare Conditional Images:

from controlnet_aux import OpenposeDetector

openpose = OpenposeDetector.from_pretrained("lllyasviel/ControlNet")
original_image = load_image(
    "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/person.png"
)
openpose_image = openpose(original_image)

prompts=""

Conditional Generation with ControlNet and PAG:

output = pipe(
    prompts,
    image=openpose_image,
    num_inference_steps=50,
    guidance_scale=0.0,
    pag_scale=4.0,
    pag_applied_layers_index=["m0"]
).images[0]

Parameters

guidance_scale : gudiance scale of CFG (ex: 7.5)

pag_scale : gudiance scale of PAG (ex: 4.0)

pag_applied_layers_index : index of the layer to apply perturbation (ex: ['m0'])