| from PIL import Image, ImageOps
|
| import numpy as np
|
| import torch
|
| MAX_RESOLUTION = 32768
|
|
|
|
|
| class ImageSimpleResize:
|
| def __init__(self):
|
| pass
|
|
|
| @classmethod
|
| def INPUT_TYPES(s):
|
| return {
|
| "required": {
|
| "image": ("IMAGE",),
|
| "size": (
|
| "INT",
|
| {"default": 512, "min": 16, "max": MAX_RESOLUTION, "step": 1},
|
| ),
|
| "edge": (["largest", "smallest", "all", "width", "height"],),
|
| },
|
| "optional": {
|
| "size_override": ("INT", {"forceInput": True}),
|
| "vae": ("VAE",),
|
| },
|
| }
|
|
|
| RETURN_TYPES = ("IMAGE", "LATENT")
|
| OUTPUT_NODE = False
|
| FUNCTION = "imagesimpleresize"
|
| CATEGORY = "Chibi-Nodes/Image"
|
|
|
| def imagesimpleresize(self, image, size, edge, size_override=None, vae=None):
|
| if size_override:
|
| size = size_override
|
|
|
| width = image.shape[2]
|
| height = image.shape[1]
|
| ratio = height / width
|
| image = Image.fromarray(
|
| np.clip(255.0 * image[0].cpu().numpy(), 0, 255).astype(np.uint8)
|
| )
|
|
|
| if edge == "largest":
|
| if width > height:
|
| if size < width:
|
| image = ImageOps.contain(
|
| image, (size, MAX_RESOLUTION), Image.LANCZOS
|
| )
|
| else:
|
| image = image.resize(
|
| (round(size), round(size * ratio)), Image.LANCZOS
|
| )
|
| if width < height:
|
| if size < height:
|
| image = ImageOps.contain(
|
| image, (MAX_RESOLUTION, size), Image.LANCZOS
|
| )
|
| else:
|
| image = image.resize(
|
| (round(size / ratio), round(size)), Image.LANCZOS
|
| )
|
| if width == height:
|
| if size < width:
|
| image = ImageOps.contain(
|
| image, (size, size), Image.LANCZOS)
|
| else:
|
| image = image.resize(
|
| (round(size), round(size)), Image.LANCZOS)
|
|
|
| if edge == "smallest":
|
| if width > height:
|
| if size < height:
|
| image = ImageOps.contain(
|
| image, (MAX_RESOLUTION, size), Image.LANCZOS
|
| )
|
| else:
|
| image = image.resize(
|
| (round(size / ratio), round(size)), Image.LANCZOS
|
| )
|
| if width < height:
|
| if size < width:
|
| image = ImageOps.contain(
|
| image, (size, MAX_RESOLUTION), Image.LANCZOS
|
| )
|
| else:
|
| image = image.resize(
|
| (round(size), round(size * ratio)), Image.LANCZOS
|
| )
|
| if width == height:
|
| if size < width:
|
| image = ImageOps.contain(
|
| image, (size, size), Image.LANCZOS)
|
| else:
|
| image = image.resize(
|
| (round(size), round(size)), Image.LANCZOS)
|
|
|
| if edge == "all":
|
| image = image.resize((round(size), round(size)), Image.LANCZOS)
|
|
|
| if edge == "width":
|
| image = image.resize((round(size), round(height)), Image.LANCZOS)
|
|
|
| if edge == "height":
|
| image = image.resize((round(width), round(size)), Image.LANCZOS)
|
|
|
| image = ImageOps.exif_transpose(image)
|
| image = image.convert("RGB")
|
| image = np.array(image).astype(np.float32) / 255.0
|
| image = torch.from_numpy(image)[None,]
|
|
|
| if vae is not None:
|
|
|
| latent = image
|
| x = (latent.shape[1] // 8) * 8
|
| y = (latent.shape[2] // 8) * 8
|
| if latent.shape[1] is not x or latent.shape[2] is not y:
|
| x_offset = (latent.shape[1] % 8) // 2
|
| y_offset = (latent.shape[2] % 8) // 2
|
| latent = latent[:, x_offset: x +
|
| x_offset, y_offset: y + y_offset, :]
|
| latent = vae.encode(latent[:, :, :, :3])
|
|
|
| return (image, {"samples": latent})
|
| else:
|
| return (
|
| image,
|
| None,
|
| )
|
|
|
