Adetailer Stable Diffusion diFFusers pipeline
This version supports fixes for input photos
pip install asdff
import torch
from asdff import AdPipeline
pipe = AdPipeline.from_pretrained("stablediffusionapi/counterfeit-v30", torch_dtype=torch.float16)
pipe.safety_checker = None
pipe.to("cuda")
common = {"prompt": "masterpiece, best quality, 1girl", "num_inference_steps": 28}
result = pipe(common=common)
images = result[0]import torch
from diffusers import DiffusionPipeline
pipe = DiffusionPipeline.from_pretrained(
"stablediffusionapi/counterfeit-v30",
torch_dtype=torch.float16,
custom_pipeline="Bingsu/adsd_pipeline"
)
pipe.safety_checker = None
pipe.to("cuda")
common = {"prompt": "masterpiece, best quality, 1girl", "num_inference_steps": 28}
result = pipe(common=common)
images = result[0]import torch
import sys
from PIL import Image
sys.path.insert(0,'/adetailer/asdff')
from asdff import AdPipeline
pipe = AdPipeline.from_pretrained("CompVis/stable-diffusion-v1-4",torch_dtype=torch.float16)
pipe.safety_checker = None
pipe.to("cuda")
img_path ='face_human.png'
res_path = 'save.png')
img=Image.open(img_path).convert("RGB")
result = pipe(common=common,images=[img])
e_time=time.time()
images.save(res_path)-
common: Mapping[str, Any] | NoneArguments used in txt2img_only
-
txt2img_only: Mapping[str, Any] | NoneArguments used in StableDiffusionPipeline
inpaint_only: Mapping[str, Any] | None
Arguments to be used only by inpaint. Arguments that overlap with common are overwritten.
strength: 0.4Used to control the degree of change in the image before and after inpainting.
StableDiffusionInpaintPipeline.call
detectors: DetectorType | Iterable[DetectorType] | None
DetectorType: Callable[[Image.Image], Optional[List[Image.Image]]]
A Callable that takes a pil Image as input and returns a list (mask) of mask images, or None.
One such Callable, a list of Callables, or None.
If None, the default_detector is used.
from asdff import AdPipeline
pipe = AdPipeline.from_pretrained(...)
pipe.default_detector
>>> <function asdff.yolo.yolo_detector(image: 'Image.Image', model_path: 'str | None' = None, confidence: 'float' = 0.3) -> 'list[Image.Image] | None'>Usage examples
from functools import partial
import torch
from asdff import AdPipeline, yolo_detector
from huggingface_hub import hf_hub_download
pipe = AdPipeline.from_pretrained("stablediffusionapi/counterfeit-v30", torch_dtype=torch.float16)
pipe.safety_checker = None
pipe.to("cuda")
person_model_path = hf_hub_download("Bingsu/adetailer", "person_yolov8s-seg.pt")
person_detector = partial(yolo_detector, model_path=person_model_path)
common = {"prompt": "masterpiece, best quality, 1girl", "num_inference_steps": 28}
result = pipe(common=common, detectors=[person_detector, pipe.default_detector])
resultmask_dilation: int, default = 4
After detecting the mask, the cv2.dilate function is applied to grow the mask area, which is the size of the kernel to be applied.
mask_blur: int, default = 4
The kernel size of the Gaussian blur to apply after dilation.
mask_padding: int, default = 32
After applying dilation, the image will be cropped by adding this value to the bbox's width and height, and then inpaint will be attempted.