nodes/lquesada-ComfyUI-Inpaint-CropAndStitch
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Sped up upscaling by enabling the selection of the rescale algorithm and switching crop default to bicubic

Browse Source
This commit is contained in:
Luis Quesada
2024-05-14 17:57:16 +02:00
parent d694780639
commit dbded659a8
2 changed files with 26 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+3 -2
View File
@@ -15,7 +15,7 @@ Check ComfyUI here: https://github.com/comfyanonymous/ComfyUI
- `context_expand_factor`: how much to grow the context area (i.e. the area for the sampling) around the original mask, as a factor, e.g. 1.1 is grow 10% of the size of the mask.
- `invert_mask`: Whether to fully invert the mask, that is, only keep what was marked, instead of removing what was marked.
- `fill_mask_holes`: Whether to fully fill any holes (small or large) in the mask, that is, mark fully enclosed areas as part of the mask.
- `blur_radius_pixels`: Whether to blur the mask, to be used in combination with `grow_mask_pixels`. Some models prefer blurred masks, some don't.
- `rescale_algorithm`: Rescale algorithm to use. bislerp is for super high quality but very slow, recommended for stich. bicubic is high quality and faster, recommended for crop.
- `mode`: Free size or Forced size.
- Free size uses `internal_rescale_factor` to optionally rescale the content before sampling and eventually scale back before stitching, and `padding` to align to standard sizes.
- Forced size uses `force_size` and upscales the content to take that size before sampling, then downscales before stitching back. Use forced size e.g. for SDXL.
@@ -64,7 +64,8 @@ If you want to inpaint with SDXL, use forced size = 1024.
# Changelog
## Upcoming!
- Some JavaScript to hide unused fields depending on the selected mode.
- Faster upscaling.
## 2024-05-14
- Enabled selecting rescaling algorithm and made bicubic the default for crop, which significantly speeds up the process.
## 2024-05-13
- Switched from adjust_to_preferred_sizes to modes: free size and forced size. Forced scales the section rather than growing the context area to fit preferred_sizes, to be used to e.g. force 1024x1024 for inpainting.
- Enable internal_upscale_factor to be lower than 1 (that is, downscale), which can be used to avoid the double head issue in some models.
inpaint_cropandstitch.py
+23 -10
View File
@@ -5,6 +5,17 @@ import numpy as np
import torch
from scipy.ndimage import gaussian_filter, grey_dilation, binary_fill_holes, binary_closing
def rescale(samples, width, height, algorithm):
if algorithm == "nearest":
return torch.nn.functional.interpolate(samples, size=(height, width), mode="nearest")
elif algorithm == "bilinear":
return torch.nn.functional.interpolate(samples, size=(height, width), mode="bilinear")
elif algorithm == "bicubic":
return torch.nn.functional.interpolate(samples, size=(height, width), mode="bicubic")
elif algorithm == "bislerp":
return comfy.utils.bislerp(samples, width, height)
return samples
class InpaintCrop:
"""
ComfyUI-InpaintCropAndStitch
@@ -25,6 +36,7 @@ class InpaintCrop:
"context_expand_factor": ("FLOAT", {"default": 1.01, "min": 1.0, "max": 100.0, "step": 0.01}),
"invert_mask": ("BOOLEAN", {"default": False}),
"fill_mask_holes": ("BOOLEAN", {"default": True}),
"rescale_algorithm": (["nearest", "bilinear", "bicubic", "bislerp"], {"default": "bicubic"}),
"mode": (["free size", "forced size"], {"default": "free size"}),
"force_size": ([512, 768, 1024, 1344, 2048, 4096, 8192], {"default": 1024}),
"rescale_factor": ("FLOAT", {"default": 1.00, "min": 0.01, "max": 100.0, "step": 0.01}),
@@ -42,7 +54,7 @@ class InpaintCrop:
FUNCTION = "inpaint_crop"
def adjust_to_square(self, x_min, x_max, y_min, y_max, width, height, target_size = None):
def adjust_to_square(self, x_min, x_max, y_min, y_max, width, height, target_size=None):
if target_size is None:
x_size = x_max - x_min + 1
y_size = y_max - y_min + 1
@@ -104,7 +116,7 @@ class InpaintCrop:
return new_min_val, new_max_val
# Parts of this function are from KJNodes: https://github.com/kijai/ComfyUI-KJNodes
def inpaint_crop(self, image, mask, context_expand_pixels, context_expand_factor, invert_mask, fill_mask_holes, mode, force_size, rescale_factor, padding, optional_context_mask = None):
def inpaint_crop(self, image, mask, context_expand_pixels, context_expand_factor, invert_mask, fill_mask_holes, mode, rescale_algorithm, force_size, rescale_factor, padding, optional_context_mask=None):
original_image = image
original_mask = mask
original_width = image.shape[2]
@@ -191,7 +203,7 @@ class InpaintCrop:
width = math.floor(samples.shape[3] * upscale_factor)
height = math.floor(samples.shape[2] * upscale_factor)
samples = comfy.utils.bislerp(samples, width, height)
samples = rescale(samples, width, height, rescale_algorithm)
effective_upscale_factor_x = float(width)/float(original_width)
effective_upscale_factor_y = float(height)/float(original_height)
samples = samples.movedim(1, -1)
@@ -199,7 +211,7 @@ class InpaintCrop:
samples = mask
samples = samples.unsqueeze(1)
samples = comfy.utils.bislerp(samples, width, height)
samples = rescale(samples, width, height, rescale_algorithm)
samples = samples.squeeze(1)
mask = samples
@@ -219,7 +231,7 @@ class InpaintCrop:
width = math.floor(samples.shape[3] * rescale_factor)
height = math.floor(samples.shape[2] * rescale_factor)
samples = comfy.utils.bislerp(samples, width, height)
samples = rescale(samples, width, height, rescale_algorithm)
effective_upscale_factor_x = float(width)/float(original_width)
effective_upscale_factor_y = float(height)/float(original_height)
samples = samples.movedim(1, -1)
@@ -227,7 +239,7 @@ class InpaintCrop:
samples = mask
samples = samples.unsqueeze(1)
samples = comfy.utils.bislerp(samples, width, height)
samples = rescale(samples, width, height, rescale_algorithm)
samples = samples.squeeze(1)
mask = samples
@@ -269,6 +281,7 @@ class InpaintStitch:
"required": {
"stitch": ("STITCH",),
"inpainted_image": ("IMAGE",),
"rescale_algorithm": (["nearest", "bilinear", "bicubic", "bislerp"], {"default": "bislerp"}),
}
}
@@ -280,7 +293,7 @@ class InpaintStitch:
FUNCTION = "inpaint_stitch"
# This function is from comfy_extras: https://github.com/comfyanonymous/ComfyUI
def composite(self, destination, source, x, y, mask = None, multiplier = 8, resize_source = False):
def composite(self, destination, source, x, y, mask=None, multiplier=8, resize_source=False):
source = source.to(destination.device)
if resize_source:
source = torch.nn.functional.interpolate(source, size=(destination.shape[2], destination.shape[3]), mode="bilinear")
@@ -314,7 +327,7 @@ class InpaintStitch:
destination[:, :, top:bottom, left:right] = source_portion + destination_portion
return destination
def inpaint_stitch(self, stitch, inpainted_image):
def inpaint_stitch(self, stitch, inpainted_image, rescale_algorithm):
original_image = stitch['original_image']
cropped_mask = stitch['cropped_mask']
x = stitch['x']
@@ -331,12 +344,12 @@ class InpaintStitch:
height = round(float(inpaint_height)/stitch['rescale_y'])
x = round(float(x)/stitch['rescale_x'])
y = round(float(y)/stitch['rescale_y'])
samples = comfy.utils.bislerp(samples, width, height)
samples = rescale(samples, width, height, rescale_algorithm)
inpainted_image = samples.movedim(1, -1)
samples = cropped_mask.movedim(-1, 1)
samples = samples.unsqueeze(0)
samples = comfy.utils.bislerp(samples, width, height)
samples = rescale(samples, width, height, rescale_algorithm)
samples = samples.squeeze(0)
cropped_mask = samples.movedim(1, -1)