Mirrored_Repos/InvokeAI
1
0
Fork 0
mirror of https://github.com/invoke-ai/InvokeAI synced 2024-08-30 20:32:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fixes inpainting not doing img2img when no mask

Browse Source
This commit is contained in:
psychedelicious 2022-11-14 12:53:45 +11:00 committed by blessedcoolant
parent 82f6402d04
commit d7884432c9
2 changed files with 53 additions and 78 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

99
backend/invoke_ai_web_server.py
View File

@ -22,8 +22,8 @@ from ldm.invoke.pngwriter import PngWriter, retrieve_metadata
from ldm.invoke.prompt_parser import split_weighted_subprompts from ldm.invoke.prompt_parser import split_weighted_subprompts
from backend.modules.parameters import parameters_to_command from backend.modules.parameters import parameters_to_command
from backend.modules.get_outpainting_generation_mode import ( from backend.modules.get_canvas_generation_mode import (
get_outpainting_generation_mode, get_canvas_generation_mode,
) )
# Loading Arguments # Loading Arguments
@ -606,7 +606,10 @@ class InvokeAIWebServer:
""" """
Prepare for generation based on generation_mode Prepare for generation based on generation_mode
""" """
if generation_parameters["generation_mode"] == "outpainting": if generation_parameters["generation_mode"] in [
"outpainting",
"inpainting",
]:
""" """
generation_parameters["init_img"] is a base64 image generation_parameters["init_img"] is a base64 image
generation_parameters["init_mask"] is a base64 image generation_parameters["init_mask"] is a base64 image
@ -617,19 +620,15 @@ class InvokeAIWebServer:
truncated_outpaint_image_b64 = generation_parameters["init_img"][:64] truncated_outpaint_image_b64 = generation_parameters["init_img"][:64]
truncated_outpaint_mask_b64 = generation_parameters["init_mask"][:64] truncated_outpaint_mask_b64 = generation_parameters["init_mask"][:64]
outpaint_image = dataURL_to_image( init_img_url = generation_parameters["init_img"]
original_bounding_box = generation_parameters["bounding_box"].copy()
if generation_parameters["generation_mode"] == "outpainting":
initial_image = dataURL_to_image(
generation_parameters["init_img"] generation_parameters["init_img"]
).convert("RGBA") ).convert("RGBA")
# Convert mask dataURL to an image and convert to greyscale
outpaint_mask = dataURL_to_image(
generation_parameters["init_mask"]
).convert("L")
actual_generation_mode = get_outpainting_generation_mode(
outpaint_image, outpaint_mask
)
""" """
The outpaint image and mask are pre-cropped by the UI, so the bounding box we pass The outpaint image and mask are pre-cropped by the UI, so the bounding box we pass
to the generator should be: to the generator should be:
@ -639,24 +638,42 @@ class InvokeAIWebServer:
"width": original_bounding_box["width"], "width": original_bounding_box["width"],
"height": original_bounding_box["height"] "height": original_bounding_box["height"]
} }
Save the original bounding box, we need to give it back to the UI when finished,
because the UI needs to know where to put the inpainted image on the canvas.
""" """
original_bounding_box = generation_parameters["bounding_box"].copy()
generation_parameters["bounding_box"]["x"] = 0 generation_parameters["bounding_box"]["x"] = 0
generation_parameters["bounding_box"]["y"] = 0 generation_parameters["bounding_box"]["y"] = 0
elif generation_parameters["generation_mode"] == "inpainting":
init_img_path = self.get_image_path_from_url(init_img_url)
initial_image = Image.open(init_img_path)
"""
For inpainting, only the mask is pre-cropped by the UI, so we need to crop out a copy
of the region of the image to be inpainted to match the size of the mask image.
"""
initial_image = copy_image_from_bounding_box(
initial_image, **generation_parameters["bounding_box"]
)
# Convert mask dataURL to an image and convert to greyscale
mask_image = dataURL_to_image(
generation_parameters["init_mask"]
).convert("L")
actual_generation_mode = get_canvas_generation_mode(
initial_image, mask_image
)
print(initial_image, mask_image)
""" """
Apply the mask to the init image, creating a "mask" image with Apply the mask to the init image, creating a "mask" image with
transparency where inpainting should occur. This is the kind of transparency where inpainting should occur. This is the kind of
mask that prompt2image() needs. mask that prompt2image() needs.
""" """
alpha_mask = outpaint_image.copy() alpha_mask = initial_image.copy()
alpha_mask.putalpha(outpaint_mask) alpha_mask.putalpha(mask_image)
generation_parameters["init_img"] = outpaint_image generation_parameters["init_img"] = initial_image
generation_parameters["init_mask"] = alpha_mask generation_parameters["init_mask"] = alpha_mask
# Remove the unneeded parameters for whichever mode we are doing # Remove the unneeded parameters for whichever mode we are doing
@ -691,48 +708,6 @@ class InvokeAIWebServer:
generation_parameters.pop("tile_size", None) generation_parameters.pop("tile_size", None)
generation_parameters.pop("force_outpaint", None) generation_parameters.pop("force_outpaint", None)
elif generation_parameters["generation_mode"] == "inpainting":
"""
generation_parameters["init_img"] is a url
generation_parameters["init_mask"] is a base64 image
So we need to convert each into a PIL Image.
"""
truncated_outpaint_image_b64 = generation_parameters["init_img"][:64]
truncated_outpaint_mask_b64 = generation_parameters["init_mask"][:64]
init_img_url = generation_parameters["init_img"]
init_mask_url = generation_parameters["init_mask"]
init_img_path = self.get_image_path_from_url(init_img_url)
original_image = Image.open(init_img_path)
rgba_image = original_image.convert("RGBA")
# copy a region from it which we will inpaint
cropped_init_image = copy_image_from_bounding_box(
rgba_image, **generation_parameters["bounding_box"]
)
original_bounding_box = generation_parameters["bounding_box"].copy()
generation_parameters["init_img"] = cropped_init_image
# Convert mask dataURL to an image and convert to greyscale
mask_image = dataURL_to_image(
generation_parameters["init_mask"]
).convert("L")
"""
Apply the mask to the init image, creating a "mask" image with
transparency where inpainting should occur. This is the kind of
mask that prompt2image() needs.
"""
alpha_mask = cropped_init_image.copy()
alpha_mask.putalpha(mask_image)
generation_parameters["init_mask"] = alpha_mask
elif generation_parameters["generation_mode"] == "img2img": elif generation_parameters["generation_mode"] == "img2img":
init_img_url = generation_parameters["init_img"] init_img_url = generation_parameters["init_img"]
init_img_path = self.get_image_path_from_url(init_img_url) init_img_path = self.get_image_path_from_url(init_img_url)

14
backend/modules/get_outpainting_generation_mode.py → backend/modules/get_canvas_generation_mode.py
View File

@ -21,7 +21,7 @@ def check_for_any_transparency(img: Union[ImageType, str]) -> bool:
return False return False
def get_outpainting_generation_mode( def get_canvas_generation_mode(
init_img: Union[ImageType, str], init_mask: Union[ImageType, str] init_img: Union[ImageType, str], init_mask: Union[ImageType, str]
) -> Literal["txt2img", "outpainting", "inpainting", "img2img",]: ) -> Literal["txt2img", "outpainting", "inpainting", "img2img",]:
if type(init_img) is str: if type(init_img) is str:
@ -80,36 +80,36 @@ def main():
print( print(
"OPAQUE IMAGE, NO MASK, expect img2img, got ", "OPAQUE IMAGE, NO MASK, expect img2img, got ",
get_outpainting_generation_mode(init_img_opaque, init_mask_no_mask), get_canvas_generation_mode(init_img_opaque, init_mask_no_mask),
) )
print( print(
"IMAGE WITH TRANSPARENCY, NO MASK, expect outpainting, got ", "IMAGE WITH TRANSPARENCY, NO MASK, expect outpainting, got ",
get_outpainting_generation_mode( get_canvas_generation_mode(
init_img_partial_transparency, init_mask_no_mask init_img_partial_transparency, init_mask_no_mask
), ),
) )
print( print(
"FULLY TRANSPARENT IMAGE NO MASK, expect txt2img, got ", "FULLY TRANSPARENT IMAGE NO MASK, expect txt2img, got ",
get_outpainting_generation_mode(init_img_full_transparency, init_mask_no_mask), get_canvas_generation_mode(init_img_full_transparency, init_mask_no_mask),
) )
print( print(
"OPAQUE IMAGE, WITH MASK, expect inpainting, got ", "OPAQUE IMAGE, WITH MASK, expect inpainting, got ",
get_outpainting_generation_mode(init_img_opaque, init_mask_has_mask), get_canvas_generation_mode(init_img_opaque, init_mask_has_mask),
) )
print( print(
"IMAGE WITH TRANSPARENCY, WITH MASK, expect outpainting, got ", "IMAGE WITH TRANSPARENCY, WITH MASK, expect outpainting, got ",
get_outpainting_generation_mode( get_canvas_generation_mode(
init_img_partial_transparency, init_mask_has_mask init_img_partial_transparency, init_mask_has_mask
), ),
) )
print( print(
"FULLY TRANSPARENT IMAGE WITH MASK, expect txt2img, got ", "FULLY TRANSPARENT IMAGE WITH MASK, expect txt2img, got ",
get_outpainting_generation_mode(init_img_full_transparency, init_mask_has_mask), get_canvas_generation_mode(init_img_full_transparency, init_mask_has_mask),
) )