up v4

app.py

@@ -1,7 +1,59 @@
+# import gradio as gr
+#
+# def greet(name):
+#     return "V5 Hello " + name + "!!"
+#
+# iface = gr.Interface(
+#     fn=greet,
+#     inputs="text",
+#     outputs="text",
+#     title="MB TEST 1",
+# )
+# iface.launch(share=True)
+
 import gradio as gr
+from models import make_inpainting
+import io
+from PIL import Image
+import numpy as np
+
+# from transformers import pipeline
+#
+# pipeline = pipeline(task="image-classification", model="julien-c/hotdog-not-hotdog")
+
+def image_to_byte_array(image: Image) -> bytes:
+    # BytesIO is a fake file stored in memory
+    imgByteArr = io.BytesIO()
+    # image.save expects a file as a argument, passing a bytes io ins
+    image.save(imgByteArr, format='png')  # image.format
+    # Turn the BytesIO object back into a bytes object
+    imgByteArr = imgByteArr.getvalue()
+    return imgByteArr
+
+def predict(input_img1,input_img2):
+
+    # image = Image.open(requests.get("https://applydesignblobs-chh5aahjdzh0cnew.z01.azurefd.net/spaceimages/org_sqr_7fee0869-3187-4363-b5fb-5233e943649d.png", stream=True).raw)
+    # mask = Image.open(requests.get("https://applydesign.blob.core.windows.net/spaceimages/mask_e85b1585-8.png", stream=True).raw)
+
+    result_image = make_inpainting(positive_prompt='test1',
+                                   image=image_to_byte_array(input_img1),
+                                   mask_image=np.array(input_img2),
+                                   negative_prompt="xxx",
+                                   )
+
+
+    # predictions = pipeline(input_img1)
+    return input_img1
+
+gradio_app = gr.Interface(
+    predict,
+    inputs=[gr.Image(label="img", sources=['upload', 'webcam'], type="pil"),
+            gr.Image(label="mask", sources=['upload', 'webcam'], type="pil")
+            ],
+    outputs= gr.Image(label="resp"),
+    title="rem fur 1",
+)
+
 
-def greet(name):
-    return "V3 Hello " + name + "!!"
+gradio_app.launch(share=True)
 
-iface = gr.Interface(fn=greet, inputs="text", outputs="text")
-iface.launch(share=True)

colors.py

@@ -0,0 +1,343 @@
+"""Color mappings"""
+from typing import List, Dict
+
+TRIVIA = {
+    "#B47878": "building;edifice",
+    "#06E6E6": "sky",
+    "#04C803": "tree",
+    "#8C8C8C": "road;route",
+    "#04FA07": "grass",
+    "#96053D": "person;individual;someone;somebody;mortal;soul",
+    "#CCFF04": "plant;flora;plant;life",
+    "#787846": "earth;ground",
+    "#FF09E0": "house",
+    "#0066C8": "car;auto;automobile;machine;motorcar",
+    "#3DE6FA": "water",
+    "#FF3D06": "railing;rail",
+    "#FF5C00": "arcade;machine",
+    "#FFE000": "stairs;steps",
+    "#00F5FF": "fan",
+    "#FF008F": "step;stair",
+    "#1F00FF": "stairway;staircase",
+    "#FFD600": "radiator",
+}
+
+OBJECTS = {
+    "#CC05FF": "bed",
+    "#FF0633": "painting;picture",
+    "#DCDCDC": "mirror",
+    "#00FF14": "box",
+    "#FF0000": "flower",
+    "#FFA300": "book",
+    "#00FFC2": "television;television;receiver;television;set;tv;tv;set;idiot;box;boob;tube;telly;goggle;box",
+    "#F500FF": "pot;flowerpot",
+    "#00FFCC": "vase",
+    "#29FF00": "tray",
+    "#8FFF00": "poster;posting;placard;notice;bill;card",
+    "#5CFF00": "basket;handbasket",
+    "#00ADFF": "screen;door;screen",
+}
+
+
+SITTING = {
+    "#0B66FF": "sofa;couch;lounge",
+    "#CC4603": "chair",
+    "#07FFE0": "seat",
+    "#08FFD6": "armchair",
+    "#FFC207": "cushion",
+    "#00EBFF": "pillow",
+    "#00D6FF": "stool",
+    "#1400FF": "blanket;cover",
+    "#0A00FF": "swivel;chair",
+    "#FF9900": "ottoman;pouf;pouffe;puff;hassock",
+}
+
+LIGHTING = {
+    "#E0FF08": "lamp",
+    "#FFAD00": "light;light;source",
+    "#001FFF": "chandelier;pendant;pendent",
+}
+
+TABLES = {
+    "#FF0652": "table",
+    "#0AFF47": "desk",
+}
+
+CLOSETS = {
+    "#E005FF": "cabinet",
+    "#FF0747": "shelf",
+    "#07FFFF": "wardrobe;closet;press",
+    "#0633FF": "chest;of;drawers;chest;bureau;dresser",
+    "#0000FF": "case;display;case;showcase;vitrine",
+}
+
+
+BATHROOM = {
+    "#6608FF": "bathtub;bathing;tub;bath;tub",
+    "#00FF85": "toilet;can;commode;crapper;pot;potty;stool;throne",
+    "#0085FF": "shower",
+    "#FF0066": "towel",
+}
+
+WINDOWS = {
+    "#FF3307": "curtain;drape;drapery;mantle;pall",
+    "#E6E6E6": "windowpane;window",
+    "#00FF3D": "awning;sunshade;sunblind",
+    "#003DFF": "blind;screen",
+}
+
+FLOOR = {
+    "#FF095C": "rug;carpet;carpeting",
+    "#503232": "floor;flooring",
+}
+
+INTERIOR = {
+    "#787878": "wall",
+    "#787850": "ceiling",
+    "#08FF33": "door;double;door",
+}
+
+KITCHEN = {
+    "#00FF29": "kitchen;island",
+    "#14FF00": "refrigerator;icebox",
+    "#00A3FF": "sink",
+    "#EB0CFF": "counter",
+    "#D6FF00": "dishwasher;dish;washer;dishwashing;machine",
+    "#FF00EB": "microwave;microwave;oven",
+    "#47FF00": "oven",
+    "#66FF00": "clock",
+    "#00FFB8": "plate",
+    "#19C2C2": "glass;drinking;glass",
+    "#00FF99": "bar",
+    "#00FF0A": "bottle",
+    "#FF7000": "buffet;counter;sideboard",
+    "#B800FF": "washer;automatic;washer;washing;machine",
+    "#00FF70": "coffee;table;cocktail;table",
+    "#008FFF": "countertop",
+    "#33FF00": "stove;kitchen;stove;range;kitchen;range;cooking;stove",
+}
+
+LIVINGROOM = {
+    "#FA0A0F": "fireplace;hearth;open;fireplace",
+    "#FF4700": "pool;table;billiard;table;snooker;table",
+}
+
+OFFICE = {
+    "#00FFAD": "computer;computing;machine;computing;device;data;processor;electronic;computer;information;processing;system",
+    "#00FFF5": "bookcase",
+    "#0633FF": "chest;of;drawers;chest;bureau;dresser",
+    "#005CFF": "monitor;monitoring;device",
+}
+
+
+COLOR_MAPPING_CATEGORY_ = {
+ 'keep background': {'#FFFFFF': 'background'},   
+ 'trivia': TRIVIA,
+ 'objects': OBJECTS,
+ 'sitting': SITTING,
+ 'lighting': LIGHTING,
+ 'tables': TABLES,
+ 'closets': CLOSETS,
+ 'bathroom': BATHROOM,
+ 'windows': WINDOWS,
+ 'floor': FLOOR,
+ 'interior': INTERIOR,
+ 'kitchen': KITCHEN,
+ 'livingroom': LIVINGROOM,
+ 'office': OFFICE}
+
+
+COLOR_MAPPING_ = {
+    '#FFFFFF': 'background',
+    "#787878": "wall",
+    "#B47878": "building;edifice",
+    "#06E6E6": "sky",
+    "#503232": "floor;flooring",
+    "#04C803": "tree",
+    "#787850": "ceiling",
+    "#8C8C8C": "road;route",
+    "#CC05FF": "bed",
+    "#E6E6E6": "windowpane;window",
+    "#04FA07": "grass",
+    "#E005FF": "cabinet",
+    "#EBFF07": "sidewalk;pavement",
+    "#96053D": "person;individual;someone;somebody;mortal;soul",
+    "#787846": "earth;ground",
+    "#08FF33": "door;double;door",
+    "#FF0652": "table",
+    "#8FFF8C": "mountain;mount",
+    "#CCFF04": "plant;flora;plant;life",
+    "#FF3307": "curtain;drape;drapery;mantle;pall",
+    "#CC4603": "chair",
+    "#0066C8": "car;auto;automobile;machine;motorcar",
+    "#3DE6FA": "water",
+    "#FF0633": "painting;picture",
+    "#0B66FF": "sofa;couch;lounge",
+    "#FF0747": "shelf",
+    "#FF09E0": "house",
+    "#0907E6": "sea",
+    "#DCDCDC": "mirror",
+    "#FF095C": "rug;carpet;carpeting",
+    "#7009FF": "field",
+    "#08FFD6": "armchair",
+    "#07FFE0": "seat",
+    "#FFB806": "fence;fencing",
+    "#0AFF47": "desk",
+    "#FF290A": "rock;stone",
+    "#07FFFF": "wardrobe;closet;press",
+    "#E0FF08": "lamp",
+    "#6608FF": "bathtub;bathing;tub;bath;tub",
+    "#FF3D06": "railing;rail",
+    "#FFC207": "cushion",
+    "#FF7A08": "base;pedestal;stand",
+    "#00FF14": "box",
+    "#FF0829": "column;pillar",
+    "#FF0599": "signboard;sign",
+    "#0633FF": "chest;of;drawers;chest;bureau;dresser",
+    "#EB0CFF": "counter",
+    "#A09614": "sand",
+    "#00A3FF": "sink",
+    "#8C8C8C": "skyscraper",
+    "#FA0A0F": "fireplace;hearth;open;fireplace",
+    "#14FF00": "refrigerator;icebox",
+    "#1FFF00": "grandstand;covered;stand",
+    "#FF1F00": "path",
+    "#FFE000": "stairs;steps",
+    "#99FF00": "runway",
+    "#0000FF": "case;display;case;showcase;vitrine",
+    "#FF4700": "pool;table;billiard;table;snooker;table",
+    "#00EBFF": "pillow",
+    "#00ADFF": "screen;door;screen",
+    "#1F00FF": "stairway;staircase",
+    "#0BC8C8": "river",
+    "#FF5200": "bridge;span",
+    "#00FFF5": "bookcase",
+    "#003DFF": "blind;screen",
+    "#00FF70": "coffee;table;cocktail;table",
+    "#00FF85": "toilet;can;commode;crapper;pot;potty;stool;throne",
+    "#FF0000": "flower",
+    "#FFA300": "book",
+    "#FF6600": "hill",
+    "#C2FF00": "bench",
+    "#008FFF": "countertop",
+    "#33FF00": "stove;kitchen;stove;range;kitchen;range;cooking;stove",
+    "#0052FF": "palm;palm;tree",
+    "#00FF29": "kitchen;island",
+    "#00FFAD": "computer;computing;machine;computing;device;data;processor;electronic;computer;information;processing;system",
+    "#0A00FF": "swivel;chair",
+    "#ADFF00": "boat",
+    "#00FF99": "bar",
+    "#FF5C00": "arcade;machine",
+    "#FF00FF": "hovel;hut;hutch;shack;shanty",
+    "#FF00F5": "bus;autobus;coach;charabanc;double-decker;jitney;motorbus;motorcoach;omnibus;passenger;vehicle",
+    "#FF0066": "towel",
+    "#FFAD00": "light;light;source",
+    "#FF0014": "truck;motortruck",
+    "#FFB8B8": "tower",
+    "#001FFF": "chandelier;pendant;pendent",
+    "#00FF3D": "awning;sunshade;sunblind",
+    "#0047FF": "streetlight;street;lamp",
+    "#FF00CC": "booth;cubicle;stall;kiosk",
+    "#00FFC2": "television;television;receiver;television;set;tv;tv;set;idiot;box;boob;tube;telly;goggle;box",
+    "#00FF52": "airplane;aeroplane;plane",
+    "#000AFF": "dirt;track",
+    "#0070FF": "apparel;wearing;apparel;dress;clothes",
+    "#3300FF": "pole",
+    "#00C2FF": "land;ground;soil",
+    "#007AFF": "bannister;banister;balustrade;balusters;handrail",
+    "#00FFA3": "escalator;moving;staircase;moving;stairway",
+    "#FF9900": "ottoman;pouf;pouffe;puff;hassock",
+    "#00FF0A": "bottle",
+    "#FF7000": "buffet;counter;sideboard",
+    "#8FFF00": "poster;posting;placard;notice;bill;card",
+    "#5200FF": "stage",
+    "#A3FF00": "van",
+    "#FFEB00": "ship",
+    "#08B8AA": "fountain",
+    "#8500FF": "conveyer;belt;conveyor;belt;conveyer;conveyor;transporter",
+    "#00FF5C": "canopy",
+    "#B800FF": "washer;automatic;washer;washing;machine",
+    "#FF001F": "plaything;toy",
+    "#00B8FF": "swimming;pool;swimming;bath;natatorium",
+    "#00D6FF": "stool",
+    "#FF0070": "barrel;cask",
+    "#5CFF00": "basket;handbasket",
+    "#00E0FF": "waterfall;falls",
+    "#70E0FF": "tent;collapsible;shelter",
+    "#46B8A0": "bag",
+    "#A300FF": "minibike;motorbike",
+    "#9900FF": "cradle",
+    "#47FF00": "oven",
+    "#FF00A3": "ball",
+    "#FFCC00": "food;solid;food",
+    "#FF008F": "step;stair",
+    "#00FFEB": "tank;storage;tank",
+    "#85FF00": "trade;name;brand;name;brand;marque",
+    "#FF00EB": "microwave;microwave;oven",
+    "#F500FF": "pot;flowerpot",
+    "#FF007A": "animal;animate;being;beast;brute;creature;fauna",
+    "#FFF500": "bicycle;bike;wheel;cycle",
+    "#0ABED4": "lake",
+    "#D6FF00": "dishwasher;dish;washer;dishwashing;machine",
+    "#00CCFF": "screen;silver;screen;projection;screen",
+    "#1400FF": "blanket;cover",
+    "#FFFF00": "sculpture",
+    "#0099FF": "hood;exhaust;hood",
+    "#0029FF": "sconce",
+    "#00FFCC": "vase",
+    "#2900FF": "traffic;light;traffic;signal;stoplight",
+    "#29FF00": "tray",
+    "#AD00FF": "ashcan;trash;can;garbage;can;wastebin;ash;bin;ash-bin;ashbin;dustbin;trash;barrel;trash;bin",
+    "#00F5FF": "fan",
+    "#4700FF": "pier;wharf;wharfage;dock",
+    "#7A00FF": "crt;screen",
+    "#00FFB8": "plate",
+    "#005CFF": "monitor;monitoring;device",
+    "#B8FF00": "bulletin;board;notice;board",
+    "#0085FF": "shower",
+    "#FFD600": "radiator",
+    "#19C2C2": "glass;drinking;glass",
+    "#66FF00": "clock",
+    "#5C00FF": "flag",
+}
+
+def ade_palette() -> List[List[int]]:
+    """ADE20K palette that maps each class to RGB values."""
+    return [[120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50],
+            [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255],
+            [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7],
+            [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82],
+            [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3],
+            [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255],
+            [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220],
+            [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224],
+            [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255],
+            [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7],
+            [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153],
+            [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255],
+            [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0],
+            [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255],
+            [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255],
+            [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255],
+            [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0],
+            [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0],
+            [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255],
+            [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255],
+            [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20],
+            [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255],
+            [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255],
+            [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255],
+            [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0],
+            [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0],
+            [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255],
+            [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112],
+            [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160],
+            [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163],
+            [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0],
+            [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0],
+            [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255],
+            [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204],
+            [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255],
+            [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255],
+            [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194],
+            [102, 255, 0], [92, 0, 255]]

config.py

@@ -0,0 +1,35 @@
+"""File with configs"""
+from palette import COLOR_MAPPING_, COLOR_MAPPING
+
+HEIGHT = 512
+WIDTH = 512
+
+def to_rgb(color: str) -> tuple:
+    """Convert hex color to rgb.
+    Args:
+        color (str): hex color
+    Returns:
+        tuple: rgb color
+    """
+    return tuple(int(color[i:i+2], 16) for i in (1, 3, 5))
+
+COLOR_NAMES = list(COLOR_MAPPING.keys())
+COLOR_RGB = [to_rgb(k) for k in COLOR_MAPPING_.keys()] + [(0, 0, 0), (255, 255, 255)]
+INVERSE_COLORS = {v: to_rgb(k) for k, v in COLOR_MAPPING_.items()}
+COLOR_MAPPING_RGB = {to_rgb(k): v for k, v in COLOR_MAPPING_.items()}
+
+def map_colors(color: str) -> str:
+    """Map color to hex value.
+    Args:
+        color (str): color name
+    Returns:
+        str: hex value
+    """
+    return COLOR_MAPPING[color]
+
+def map_colors_rgb(color: tuple) -> str:
+    return COLOR_MAPPING_RGB[color]
+
+
+POS_PROMPT = "tree, sky, cloud, scenery, outdoors, grass, flowers, sunlight, beautiful, ultra detailed beautiful landscape, architectural renderings vegetation, high res, best high quality landscape, outdoor lighting, sunshine, 4k, 8k, realistic"
+NEG_PROMPT= "lowres, deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, mutated hands and fingers, out of frame"

explanation.py

@@ -0,0 +1,51 @@
+import streamlit as st
+
+def make_inpainting_explanation():
+    with st.expander("Explanation inpainting", expanded=False):
+        st.write("In the inpainting mode, you can draw regions on the input image that you want to regenerate. "
+                "This can be useful to remove unwanted objects from the image or to improve the consistency of the image."
+                )
+        st.image("content/inpainting_sidebar.png", caption="Image before inpainting, note the ornaments on the wall", width=500)
+        st.write("You can find drawing options in the sidebar. There are two modes: freedraw and polygon. Freedraw allows the user to draw with a pencil of a certain width. "
+                 "Polygon allows the user to draw a polygon by clicking on the image to add a point. The polygon is closed by right clicking.")
+
+        st.write("### Example inpainting")
+        st.write("In the example below, the ornaments on the wall are removed. The inpainting is done by drawing a mask on the image.")
+        st.image("content/inpainting_before.jpg", caption="Image before inpainting, note the ornaments on the wall")
+        st.image("content/inpainting_after.png", caption="Image before inpainting, note the ornaments on the wall")
+        
+def make_regeneration_explanation():
+    with st.expander("Explanation object regeneration"):
+        st.write("In this object regeneration mode, the model calculates which objects occur in the image. "
+                 "The user can then select which objects can be regenerated by the controlnet model by adding them in the multiselect box. "
+                 "All the object classes that are not selected will remain the same as in the original image."
+                 )
+        st.write("### Example object regeneration")
+        st.write("In the example below, the room consists of various objects such as wall, ceiling, floor, lamp, bed, ... "
+                 "In the multiselect box, all the objects except for 'lamp', 'bed and 'table' are selected to be regenerated. "
+                 )
+        st.image("content/regen_example.png", caption="Room where all concepts except for 'bed', 'lamp', 'table' are regenerated")
+
+def make_segmentation_explanation():
+    with st.expander("Segmentation mode", expanded=False):
+        st.write("In the segmentation mode, the user can use his imagination and the paint brush to place concepts in the image. "
+                 "In the left sidebar, you can first find the high level category of the concept you want to add, such as 'lighting', 'floor', .. "
+                 "After selecting the category, you can select the specific concept you want to add in the 'Choose a color' dropdown. "
+                 "This will change the color of the paint brush, which you can then use to draw on the input image. "
+                 "The model will then regenerate the image with the concepts you have drawn and leave the rest of the image unchanged. "
+                 )
+        st.image("content/sidebar segmentation.png", caption="Sidebar with segmentation options", width=300)
+        st.write("You can choose the freedraw mode which gives you a pencil of a certain (chosen) width or the polygon mode. With the polygon mode you can click to add a point to the polygon and close the polygon by right clicking. ")
+        st.write("Important: "
+                 "it's not easy to draw a good segmentation mask. This is because you need to keep in mind the perspective of the room and the exact "
+                 "shape of the object you want to draw within this perspective. Controlnet will follow your segmentation mask pretty well, so "
+                 "a non-natural object shape will sometimes result in weird outputs. However, give it a try and see what you can do! "
+                 )
+        st.image("content/segmentation window.png", caption="Example of a segmentation mask drawn on the input image to add a window to the room")
+        st.write("Tip: ")
+        st.write("In the concepts dropdown, you can select 'keep background' (which is a white color). Everything drawn in this color will use "
+                 "the original underlying segmentation mask. This can be useful to help with generating other objects, since you give the model a some "
+                 "freedom to generate outside the object borders."
+                 )
+        st.image("content/keep background 1.png", caption="Image with a poster drawn on the wall.")
+        st.image("content/keep background 2.png", caption="Image with a poster drawn on the wall surrounded by 'keep background'.")

helpers.py

@@ -0,0 +1,47 @@
+import gc
+import torch
+from scipy.signal import fftconvolve
+from PIL import Image
+import numpy as np
+
+def flush():
+    gc.collect()
+    torch.cuda.empty_cache()
+    
+
+
+def convolution(mask: Image.Image, size=9) -> Image:
+    """Method to blur the mask
+    Args:
+        mask (Image): masking image
+        size (int, optional): size of the blur. Defaults to 9.
+    Returns:
+        Image: blurred mask
+    """
+    mask = np.array(mask.convert("L"))
+    conv = np.ones((size, size)) / size**2
+    mask_blended = fftconvolve(mask, conv, 'same')
+    mask_blended = mask_blended.astype(np.uint8).copy()
+
+    border = size
+
+    # replace borders with original values
+    mask_blended[:border, :] = mask[:border, :]
+    mask_blended[-border:, :] = mask[-border:, :]
+    mask_blended[:, :border] = mask[:, :border]
+    mask_blended[:, -border:] = mask[:, -border:]
+
+    return Image.fromarray(mask_blended).convert("L")
+
+
+def postprocess_image_masking(inpainted: Image, image: Image, mask: Image) -> Image:
+    """Method to postprocess the inpainted image
+    Args:
+        inpainted (Image): inpainted image
+        image (Image): original image
+        mask (Image): mask
+    Returns:
+        Image: inpainted image
+    """
+    final_inpainted = Image.composite(inpainted.convert("RGBA"), image.convert("RGBA"), mask)
+    return final_inpainted.convert("RGB")

models.py

@@ -0,0 +1,98 @@
+"""This file contains methods for inference and image generation."""
+import logging
+from typing import List, Tuple, Dict
+
+import streamlit as st
+import torch
+import gc
+import time
+import numpy as np
+from PIL import Image
+from PIL import ImageFilter
+
+from diffusers import ControlNetModel, UniPCMultistepScheduler
+
+from config import WIDTH, HEIGHT
+from palette import ade_palette
+from stable_diffusion_controlnet_inpaint_img2img import StableDiffusionControlNetInpaintImg2ImgPipeline
+from helpers import flush, postprocess_image_masking, convolution
+from pipelines import ControlNetPipeline, SDPipeline, get_inpainting_pipeline, get_controlnet
+
+LOGGING = logging.getLogger(__name__)
+
+
+@torch.inference_mode()
+def make_image_controlnet(image: np.ndarray,
+                          mask_image: np.ndarray,
+                          controlnet_conditioning_image: np.ndarray,
+                          positive_prompt: str, negative_prompt: str,
+                          seed: int = 2356132) -> List[Image.Image]:
+    """Method to make image using controlnet
+    Args:
+        image (np.ndarray): input image
+        mask_image (np.ndarray): mask image
+        controlnet_conditioning_image (np.ndarray): conditioning image
+        positive_prompt (str): positive prompt string
+        negative_prompt (str): negative prompt string
+        seed (int, optional): seed. Defaults to 2356132.
+    Returns:
+        List[Image.Image]: list of generated images
+    """
+
+    pipe = get_controlnet()
+    flush()
+
+    image = Image.fromarray(image).convert("RGB")
+    controlnet_conditioning_image = Image.fromarray(controlnet_conditioning_image).convert("RGB")#.filter(ImageFilter.GaussianBlur(radius = 9))
+    mask_image = Image.fromarray((mask_image * 255).astype(np.uint8)).convert("RGB")
+    mask_image_postproc = convolution(mask_image)
+
+
+    st.success(f"{pipe.queue_size} images in the queue, can take up to {(pipe.queue_size+1) * 10} seconds")
+    generated_image = pipe(
+        prompt=positive_prompt,
+        negative_prompt=negative_prompt,
+        num_inference_steps=50,
+        strength=1.00,
+        guidance_scale=7.0,
+        generator=[torch.Generator(device="cuda").manual_seed(seed)],
+        image=image,
+        mask_image=mask_image,
+        controlnet_conditioning_image=controlnet_conditioning_image,
+    ).images[0]
+    generated_image = postprocess_image_masking(generated_image, image, mask_image_postproc)
+
+    return generated_image
+
+
+@torch.inference_mode()
+def make_inpainting(positive_prompt: str,
+                    image: Image,
+                    mask_image: np.ndarray,
+                    negative_prompt: str = "") -> List[Image.Image]:
+    """Method to make inpainting
+    Args:
+        positive_prompt (str): positive prompt string
+        image (Image): input image
+        mask_image (np.ndarray): mask image
+        negative_prompt (str, optional): negative prompt string. Defaults to "".
+    Returns:
+        List[Image.Image]: list of generated images
+    """
+    pipe = get_inpainting_pipeline()
+    mask_image = Image.fromarray((mask_image * 255).astype(np.uint8))
+    mask_image_postproc = convolution(mask_image)
+
+    flush()
+    st.success(f"{pipe.queue_size} images in the queue, can take up to {(pipe.queue_size+1) * 10} seconds")
+    generated_image = pipe(image=image,
+                    mask_image=mask_image,
+                    prompt=positive_prompt,
+                    negative_prompt=negative_prompt,
+                    num_inference_steps=50,
+                    height=HEIGHT,
+                    width=WIDTH,
+                    ).images[0]
+    generated_image = postprocess_image_masking(generated_image, image, mask_image_postproc)
+
+    return generated_image

palette.py

@@ -0,0 +1,38 @@
+"""This file contains color information"""
+from typing import List, Dict
+from colors import COLOR_MAPPING_, COLOR_MAPPING_CATEGORY_, ade_palette
+
+
+def convert_hex_to_rgba(hex_code: str) -> str:
+    """Convert hex code to rgba.
+    Args:
+        hex_code (str): hex string
+    Returns:
+        str: rgba string
+    """
+    hex_code = hex_code.lstrip('#')
+    return "rgba(" + str(int(hex_code[0:2], 16)) + ", " + str(int(hex_code[2:4], 16)) + ", " + str(int(hex_code[4:6], 16)) + ", 1.0)"
+
+
+def convert_dict_to_rgba(color_dict: Dict) -> Dict:
+    """Convert hex code to rgba for all elements in a dictionary.
+    Args:
+        color_dict (Dict): color dictionary
+    Returns:
+        Dict: color dictionary with rgba values
+    """
+    updated_dict = {}
+    for k, v in color_dict.items():
+        updated_dict[convert_hex_to_rgba(k)] = v
+    return updated_dict
+
+
+def convert_nested_dict_to_rgba(nested_dict):
+    updated_dict = {}
+    for k, v in nested_dict.items():
+        updated_dict[k] = convert_dict_to_rgba(v)
+    return updated_dict
+
+
+COLOR_MAPPING = convert_dict_to_rgba(COLOR_MAPPING_)
+COLOR_MAPPING_CATEGORY = convert_nested_dict_to_rgba(COLOR_MAPPING_CATEGORY_)

pipelines.py

@@ -0,0 +1,126 @@
+import logging
+from typing import List, Tuple, Dict
+
+import streamlit as st
+import torch
+import gc
+import time
+import numpy as np
+from PIL import Image
+from time import perf_counter
+from contextlib import contextmanager
+from scipy.signal import fftconvolve
+from PIL import ImageFilter
+
+from diffusers import ControlNetModel, UniPCMultistepScheduler
+from diffusers import StableDiffusionInpaintPipeline
+
+from config import WIDTH, HEIGHT
+from stable_diffusion_controlnet_inpaint_img2img import StableDiffusionControlNetInpaintImg2ImgPipeline
+from helpers import flush
+
+LOGGING = logging.getLogger(__name__)
+
+class ControlNetPipeline:
+    def __init__(self):
+        self.in_use = False
+        self.controlnet = ControlNetModel.from_pretrained(
+        "BertChristiaens/controlnet-seg-room", torch_dtype=torch.float16)
+
+        self.pipe = StableDiffusionControlNetInpaintImg2ImgPipeline.from_pretrained(
+            "runwayml/stable-diffusion-inpainting",
+            controlnet=self.controlnet,
+            safety_checker=None,
+            torch_dtype=torch.float16
+        )
+
+        self.pipe.scheduler = UniPCMultistepScheduler.from_config(self.pipe.scheduler.config)
+        self.pipe.enable_xformers_memory_efficient_attention()
+        self.pipe = self.pipe.to("cuda")
+        
+        self.waiting_queue = []
+        self.count = 0
+    
+    @property
+    def queue_size(self):
+        return len(self.waiting_queue)
+    
+    def __call__(self, **kwargs):
+        self.count += 1
+        number = self.count
+
+        self.waiting_queue.append(number)
+        
+        # wait until the next number in the queue is the current number
+        while self.waiting_queue[0] != number:
+            print(f"Wait for your turn {number} in queue {self.waiting_queue}")
+            time.sleep(0.5)
+            pass
+
+        # it's your turn, so remove the number from the queue
+        # and call the function
+        print("It's the turn of", self.count)
+        results = self.pipe(**kwargs)
+        self.waiting_queue.pop(0)
+        flush()
+        return results
+    
+class SDPipeline:
+    def __init__(self):
+        self.pipe = StableDiffusionInpaintPipeline.from_pretrained(
+            "stabilityai/stable-diffusion-2-inpainting",
+            torch_dtype=torch.float16,
+            safety_checker=None,
+        )
+
+        self.pipe.enable_xformers_memory_efficient_attention()
+        # self.pipe = self.pipe.to("cuda")
+        
+        self.waiting_queue = []
+        self.count = 0
+    
+    @property
+    def queue_size(self):
+        return len(self.waiting_queue)
+    
+    def __call__(self, **kwargs):
+        self.count += 1
+        number = self.count
+
+        self.waiting_queue.append(number)
+        
+        # wait until the next number in the queue is the current number
+        while self.waiting_queue[0] != number:
+            print(f"Wait for your turn {number} in queue {self.waiting_queue}")
+            time.sleep(0.5)
+            pass
+
+        # it's your turn, so remove the number from the queue
+        # and call the function
+        print("It's the turn of", self.count)
+        results = self.pipe(**kwargs)
+        self.waiting_queue.pop(0)
+        flush()
+        return results
+
+
+
+@st.cache_resource(max_entries=5)
+def get_controlnet():
+    """Method to load the controlnet model
+    Returns:
+        ControlNetModel: controlnet model
+    """
+    pipe = ControlNetPipeline()
+    return pipe
+
+
+
+@st.cache_resource(max_entries=5)
+def get_inpainting_pipeline():
+    """Method to load the inpainting pipeline
+    Returns:
+        StableDiffusionInpaintPipeline: inpainting pipeline
+    """
+    pipe = SDPipeline()
+    return pipe

preprocessing.py

@@ -0,0 +1,134 @@
+"""Preprocessing methods"""
+import logging
+from typing import List, Tuple
+
+import numpy as np
+from PIL import Image, ImageFilter
+import streamlit as st
+
+from config import COLOR_RGB, WIDTH, HEIGHT
+# from enhance_config import ENHANCE_SETTINGS
+
+LOGGING = logging.getLogger(__name__)
+
+
+def preprocess_seg_mask(canvas_seg, real_seg: Image.Image = None) -> Tuple[np.ndarray, np.ndarray]:
+    """Preprocess the segmentation mask.
+    Args:
+        canvas_seg: segmentation canvas
+        real_seg (Image.Image, optional): segmentation mask. Defaults to None.
+    Returns:
+        Tuple[np.ndarray, np.ndarray]: segmentation mask, segmentation mask with overlay
+    """
+    # get unique colors in the segmentation
+    image_seg = canvas_seg.image_data.copy()[:, :, :3]
+
+    # average the colors of the segmentation masks
+    average_color = np.mean(image_seg, axis=(2))
+    mask = average_color[:, :] > 0
+    if mask.sum() > 0:
+        mask = mask * 1
+
+    unique_colors = np.unique(image_seg.reshape(-1, image_seg.shape[-1]), axis=0)
+    unique_colors = [tuple(color) for color in unique_colors]
+
+    unique_colors = [color for color in unique_colors if np.sum(
+        np.all(image_seg == color, axis=-1)) > 100]
+
+    unique_colors_exact = [color for color in unique_colors if color in COLOR_RGB]
+
+    if real_seg is not None:
+        overlay_seg = np.array(real_seg)
+
+        unique_colors = np.unique(overlay_seg.reshape(-1, overlay_seg.shape[-1]), axis=0)
+        unique_colors = [tuple(color) for color in unique_colors]
+
+        for color in unique_colors_exact:
+            if color != (255, 255, 255) and color != (0, 0, 0):
+                overlay_seg[np.all(image_seg == color, axis=-1)] = color
+        image_seg = overlay_seg
+
+    return mask, image_seg
+
+
+def get_mask(image_mask: np.ndarray) -> np.ndarray:
+    """Get the mask from the segmentation mask.
+    Args:
+        image_mask (np.ndarray): segmentation mask
+    Returns:
+        np.ndarray: mask
+    """
+    # average the colors of the segmentation masks
+    average_color = np.mean(image_mask, axis=(2))
+    mask = average_color[:, :] > 0
+    if mask.sum() > 0:
+        mask = mask * 1
+    return mask
+
+
+def get_image() -> np.ndarray:
+    """Get the image from the session state.
+    Returns:
+        np.ndarray: image
+    """
+    if 'initial_image' in st.session_state and st.session_state['initial_image'] is not None:
+        initial_image = st.session_state['initial_image']
+        if isinstance(initial_image, Image.Image):
+            return np.array(initial_image.resize((WIDTH, HEIGHT)))
+        else:
+            return np.array(Image.fromarray(initial_image).resize((WIDTH, HEIGHT)))
+    else:
+        return None
+
+
+# def make_enhance_config(segmentation, objects=None):
+    """Make the enhance config for the segmentation image.
+    """
+    info = ENHANCE_SETTINGS[objects]
+
+    segmentation = np.array(segmentation)
+
+    if 'replace' in info:
+        replace_color = info['replace']
+        mask = np.zeros(segmentation.shape)
+        for color in info['colors']:
+            mask[np.all(segmentation == color, axis=-1)] = [1, 1, 1]
+            segmentation[np.all(segmentation == color, axis=-1)] = replace_color
+
+    if info['inverse'] is False:
+        mask = np.zeros(segmentation.shape)
+        for color in info['colors']:
+            mask[np.all(segmentation == color, axis=-1)] = [1, 1, 1]
+    else:
+        mask = np.ones(segmentation.shape)
+        for color in info['colors']:
+            mask[np.all(segmentation == color, axis=-1)] = [0, 0, 0]
+
+    st.session_state['positive_prompt'] = info['positive_prompt']
+    st.session_state['negative_prompt'] = info['negative_prompt']
+
+    if info['inpainting'] is True:
+        mask = mask.astype(np.uint8)
+        mask = Image.fromarray(mask)
+        mask = mask.filter(ImageFilter.GaussianBlur(radius=13))
+        mask = mask.filter(ImageFilter.MaxFilter(size=9))
+        mask = np.array(mask)
+
+        mask[mask < 0.1] = 0
+        mask[mask >= 0.1] = 1
+        mask = mask.astype(np.uint8)
+
+        conditioning = dict(
+            mask_image=mask,
+            positive_prompt=info['positive_prompt'],
+            negative_prompt=info['negative_prompt'],
+        )
+    else:
+        conditioning = dict(
+            mask_image=mask,
+            controlnet_conditioning_image=segmentation,
+            positive_prompt=info['positive_prompt'],
+            negative_prompt=info['negative_prompt'],
+            strength=info['strength']
+        )
+    return conditioning, info['inpainting']

requirements.txt

@@ -1,14 +1,14 @@
-streamlit
-streamlit-drawable-canvas
-diffusers
-xformers
-transformers
-torchvision
+streamlit==1.20.0
+streamlit-drawable-canvas==0.9.0
+diffusers==0.15.0
+xformers==0.0.16
+transformers==4.28.0
+torchvision==0.14.1
 git+https://github.com/huggingface/accelerate.git
-opencv-python-headless
-scipy
+opencv-python-headless==4.7.0.72
+scipy==1.10.0
 python-docx
-extra-streamlit-components
+extra-streamlit-components==0.1.56
 triton
-altair
+altair==4.1.0
 gradio

segmentation.py

@@ -0,0 +1,55 @@
+import logging
+from typing import List, Tuple, Dict
+
+import streamlit as st
+import torch
+import gc
+import numpy as np
+from PIL import Image
+
+from transformers import AutoImageProcessor, UperNetForSemanticSegmentation
+
+from palette import ade_palette
+
+LOGGING = logging.getLogger(__name__)
+
+
+def flush():
+    gc.collect()
+    torch.cuda.empty_cache()
+
+@st.cache_resource(max_entries=5)
+def get_segmentation_pipeline() -> Tuple[AutoImageProcessor, UperNetForSemanticSegmentation]:
+    """Method to load the segmentation pipeline
+    Returns:
+        Tuple[AutoImageProcessor, UperNetForSemanticSegmentation]: segmentation pipeline
+    """
+    image_processor = AutoImageProcessor.from_pretrained("openmmlab/upernet-convnext-small")
+    image_segmentor = UperNetForSemanticSegmentation.from_pretrained(
+        "openmmlab/upernet-convnext-small")
+    return image_processor, image_segmentor
+
+
+@torch.inference_mode()
+@torch.autocast('cuda')
+def segment_image(image: Image) -> Image:
+    """Method to segment image
+    Args:
+        image (Image): input image
+    Returns:
+        Image: segmented image
+    """
+    image_processor, image_segmentor = get_segmentation_pipeline()
+    pixel_values = image_processor(image, return_tensors="pt").pixel_values
+    with torch.no_grad():
+        outputs = image_segmentor(pixel_values)
+
+    seg = image_processor.post_process_semantic_segmentation(
+        outputs, target_sizes=[image.size[::-1]])[0]
+    color_seg = np.zeros((seg.shape[0], seg.shape[1], 3), dtype=np.uint8)
+    palette = np.array(ade_palette())
+    for label, color in enumerate(palette):
+        color_seg[seg == label, :] = color
+    color_seg = color_seg.astype(np.uint8)
+    seg_image = Image.fromarray(color_seg).convert('RGB')
+    return seg_image

stable_diffusion_controlnet_inpaint_img2img.py

@@ -0,0 +1,1112 @@
+"""This file contains the StableDiffusionControlNetInpaintImg2ImgPipeline class from the
+community pipelines from the diffusers library of HuggingFace.
+"""
+# Inspired by: https://github.com/haofanwang/ControlNet-for-Diffusers/
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Union
+
+import numpy as np
+import PIL.Image
+import torch
+import torch.nn.functional as F
+from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
+
+from diffusers import AutoencoderKL, ControlNetModel, DiffusionPipeline, UNet2DConditionModel, logging
+from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput, StableDiffusionSafetyChecker
+from diffusers.schedulers import KarrasDiffusionSchedulers
+from diffusers.utils import (
+    PIL_INTERPOLATION,
+    is_accelerate_available,
+    is_accelerate_version,
+    randn_tensor,
+    replace_example_docstring,
+)
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> import numpy as np
+        >>> import torch
+        >>> from PIL import Image
+        >>> from stable_diffusion_controlnet_inpaint_img2img import StableDiffusionControlNetInpaintImg2ImgPipeline
+        >>> from transformers import AutoImageProcessor, UperNetForSemanticSegmentation
+        >>> from diffusers import ControlNetModel, UniPCMultistepScheduler
+        >>> from diffusers.utils import load_image
+        >>> def ade_palette():
+                return [[120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50],
+                        [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255],
+                        [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7],
+                        [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82],
+                        [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3],
+                        [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255],
+                        [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220],
+                        [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224],
+                        [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255],
+                        [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7],
+                        [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153],
+                        [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255],
+                        [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0],
+                        [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255],
+                        [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255],
+                        [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255],
+                        [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0],
+                        [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0],
+                        [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255],
+                        [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255],
+                        [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20],
+                        [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255],
+                        [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255],
+                        [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255],
+                        [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0],
+                        [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0],
+                        [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255],
+                        [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112],
+                        [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160],
+                        [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163],
+                        [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0],
+                        [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0],
+                        [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255],
+                        [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204],
+                        [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255],
+                        [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255],
+                        [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194],
+                        [102, 255, 0], [92, 0, 255]]
+        >>> image_processor = AutoImageProcessor.from_pretrained("openmmlab/upernet-convnext-small")
+        >>> image_segmentor = UperNetForSemanticSegmentation.from_pretrained("openmmlab/upernet-convnext-small")
+        >>> pipe = StableDiffusionControlNetInpaintImg2ImgPipeline.from_pretrained(
+                "runwayml/stable-diffusion-inpainting", controlnet=controlnet, safety_checker=None, torch_dtype=torch.float16
+            )
+        >>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
+        >>> pipe.enable_xformers_memory_efficient_attention()
+        >>> pipe.enable_model_cpu_offload()
+        >>> def image_to_seg(image):
+                pixel_values = image_processor(image, return_tensors="pt").pixel_values
+                with torch.no_grad():
+                    outputs = image_segmentor(pixel_values)
+                seg = image_processor.post_process_semantic_segmentation(outputs, target_sizes=[image.size[::-1]])[0]
+                color_seg = np.zeros((seg.shape[0], seg.shape[1], 3), dtype=np.uint8)  # height, width, 3
+                palette = np.array(ade_palette())
+                for label, color in enumerate(palette):
+                    color_seg[seg == label, :] = color
+                color_seg = color_seg.astype(np.uint8)
+                seg_image = Image.fromarray(color_seg)
+                return seg_image
+        >>> image = load_image(
+                "https://github.com/CompVis/latent-diffusion/raw/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
+            )
+        >>> mask_image = load_image(
+                "https://github.com/CompVis/latent-diffusion/raw/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
+            )
+        >>> controlnet_conditioning_image = image_to_seg(image)
+        >>> image = pipe(
+                "Face of a yellow cat, high resolution, sitting on a park bench",
+                image,
+                mask_image,
+                controlnet_conditioning_image,
+                num_inference_steps=20,
+            ).images[0]
+        >>> image.save("out.png")
+        ```
+"""
+
+
+def prepare_image(image):
+    if isinstance(image, torch.Tensor):
+        # Batch single image
+        if image.ndim == 3:
+            image = image.unsqueeze(0)
+
+        image = image.to(dtype=torch.float32)
+    else:
+        # preprocess image
+        if isinstance(image, (PIL.Image.Image, np.ndarray)):
+            image = [image]
+
+        if isinstance(image, list) and isinstance(image[0], PIL.Image.Image):
+            image = [np.array(i.convert("RGB"))[None, :] for i in image]
+            image = np.concatenate(image, axis=0)
+        elif isinstance(image, list) and isinstance(image[0], np.ndarray):
+            image = np.concatenate([i[None, :] for i in image], axis=0)
+
+        image = image.transpose(0, 3, 1, 2)
+        image = torch.from_numpy(image).to(dtype=torch.float32) / 127.5 - 1.0
+
+    return image
+
+
+def prepare_mask_image(mask_image):
+    if isinstance(mask_image, torch.Tensor):
+        if mask_image.ndim == 2:
+            # Batch and add channel dim for single mask
+            mask_image = mask_image.unsqueeze(0).unsqueeze(0)
+        elif mask_image.ndim == 3 and mask_image.shape[0] == 1:
+            # Single mask, the 0'th dimension is considered to be
+            # the existing batch size of 1
+            mask_image = mask_image.unsqueeze(0)
+        elif mask_image.ndim == 3 and mask_image.shape[0] != 1:
+            # Batch of mask, the 0'th dimension is considered to be
+            # the batching dimension
+            mask_image = mask_image.unsqueeze(1)
+
+        # Binarize mask
+        mask_image[mask_image < 0.5] = 0
+        mask_image[mask_image >= 0.5] = 1
+    else:
+        # preprocess mask
+        if isinstance(mask_image, (PIL.Image.Image, np.ndarray)):
+            mask_image = [mask_image]
+
+        if isinstance(mask_image, list) and isinstance(mask_image[0], PIL.Image.Image):
+            mask_image = np.concatenate([np.array(m.convert("L"))[None, None, :] for m in mask_image], axis=0)
+            mask_image = mask_image.astype(np.float32) / 255.0
+        elif isinstance(mask_image, list) and isinstance(mask_image[0], np.ndarray):
+            mask_image = np.concatenate([m[None, None, :] for m in mask_image], axis=0)
+
+        mask_image[mask_image < 0.5] = 0
+        mask_image[mask_image >= 0.5] = 1
+        mask_image = torch.from_numpy(mask_image)
+
+    return mask_image
+
+
+def prepare_controlnet_conditioning_image(
+    controlnet_conditioning_image, width, height, batch_size, num_images_per_prompt, device, dtype
+):
+    if not isinstance(controlnet_conditioning_image, torch.Tensor):
+        if isinstance(controlnet_conditioning_image, PIL.Image.Image):
+            controlnet_conditioning_image = [controlnet_conditioning_image]
+
+        if isinstance(controlnet_conditioning_image[0], PIL.Image.Image):
+            controlnet_conditioning_image = [
+                np.array(i.resize((width, height), resample=PIL_INTERPOLATION["lanczos"]))[None, :]
+                for i in controlnet_conditioning_image
+            ]
+            controlnet_conditioning_image = np.concatenate(controlnet_conditioning_image, axis=0)
+            controlnet_conditioning_image = np.array(controlnet_conditioning_image).astype(np.float32) / 255.0
+            controlnet_conditioning_image = controlnet_conditioning_image.transpose(0, 3, 1, 2)
+            controlnet_conditioning_image = torch.from_numpy(controlnet_conditioning_image)
+        elif isinstance(controlnet_conditioning_image[0], torch.Tensor):
+            controlnet_conditioning_image = torch.cat(controlnet_conditioning_image, dim=0)
+
+    image_batch_size = controlnet_conditioning_image.shape[0]
+
+    if image_batch_size == 1:
+        repeat_by = batch_size
+    else:
+        # image batch size is the same as prompt batch size
+        repeat_by = num_images_per_prompt
+
+    controlnet_conditioning_image = controlnet_conditioning_image.repeat_interleave(repeat_by, dim=0)
+
+    controlnet_conditioning_image = controlnet_conditioning_image.to(device=device, dtype=dtype)
+
+    return controlnet_conditioning_image
+
+
+class StableDiffusionControlNetInpaintImg2ImgPipeline(DiffusionPipeline):
+    """
+    Inspired by: https://github.com/haofanwang/ControlNet-for-Diffusers/
+    """
+
+    _optional_components = ["safety_checker", "feature_extractor"]
+
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        unet: UNet2DConditionModel,
+        controlnet: ControlNetModel,
+        scheduler: KarrasDiffusionSchedulers,
+        safety_checker: StableDiffusionSafetyChecker,
+        feature_extractor: CLIPFeatureExtractor,
+        requires_safety_checker: bool = True,
+    ):
+        super().__init__()
+
+        if safety_checker is None and requires_safety_checker:
+            logger.warning(
+                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+                " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+            )
+
+        if safety_checker is not None and feature_extractor is None:
+            raise ValueError(
+                "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"
+                " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
+            )
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            unet=unet,
+            controlnet=controlnet,
+            scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=feature_extractor,
+        )
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.register_to_config(requires_safety_checker=requires_safety_checker)
+
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding.
+        When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several
+        steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously invoked, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_sequential_cpu_offload(self, gpu_id=0):
+        r"""
+        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
+        text_encoder, vae, controlnet, and safety checker have their state dicts saved to CPU and then are moved to a
+        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
+        Note that offloading happens on a submodule basis. Memory savings are higher than with
+        `enable_model_cpu_offload`, but performance is lower.
+        """
+        if is_accelerate_available():
+            from accelerate import cpu_offload
+        else:
+            raise ImportError("Please install accelerate via `pip install accelerate`")
+
+        device = torch.device(f"cuda:{gpu_id}")
+
+        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.controlnet]:
+            cpu_offload(cpu_offloaded_model, device)
+
+        if self.safety_checker is not None:
+            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
+
+    def enable_model_cpu_offload(self, gpu_id=0):
+        r"""
+        Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
+        to `enable_sequential_cpu_offload`, this method moves one whole model at a time to the GPU when its `forward`
+        method is called, and the model remains in GPU until the next model runs. Memory savings are lower than with
+        `enable_sequential_cpu_offload`, but performance is much better due to the iterative execution of the `unet`.
+        """
+        if is_accelerate_available() and is_accelerate_version(">=", "0.17.0.dev0"):
+            from accelerate import cpu_offload_with_hook
+        else:
+            raise ImportError("`enable_model_offload` requires `accelerate v0.17.0` or higher.")
+
+        device = torch.device(f"cuda:{gpu_id}")
+
+        hook = None
+        for cpu_offloaded_model in [self.text_encoder, self.unet, self.vae]:
+            _, hook = cpu_offload_with_hook(cpu_offloaded_model, device, prev_module_hook=hook)
+
+        if self.safety_checker is not None:
+            # the safety checker can offload the vae again
+            _, hook = cpu_offload_with_hook(self.safety_checker, device, prev_module_hook=hook)
+
+        # control net hook has be manually offloaded as it alternates with unet
+        cpu_offload_with_hook(self.controlnet, device)
+
+        # We'll offload the last model manually.
+        self.final_offload_hook = hook
+
+    @property
+    def _execution_device(self):
+        r"""
+        Returns the device on which the pipeline's models will be executed. After calling
+        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
+        hooks.
+        """
+        if not hasattr(self.unet, "_hf_hook"):
+            return self.device
+        for module in self.unet.modules():
+            if (
+                hasattr(module, "_hf_hook")
+                and hasattr(module._hf_hook, "execution_device")
+                and module._hf_hook.execution_device is not None
+            ):
+                return torch.device(module._hf_hook.execution_device)
+        return self.device
+
+    def _encode_prompt(
+        self,
+        prompt,
+        device,
+        num_images_per_prompt,
+        do_classifier_free_guidance,
+        negative_prompt=None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+    ):
+        r"""
+        Encodes the prompt into text encoder hidden states.
+        Args:
+             prompt (`str` or `List[str]`, *optional*):
+                prompt to be encoded
+            device: (`torch.device`):
+                torch device
+            num_images_per_prompt (`int`):
+                number of images that should be generated per prompt
+            do_classifier_free_guidance (`bool`):
+                whether to use classifier free guidance or not
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.
+                Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+        """
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        if prompt_embeds is None:
+            text_inputs = self.tokenizer(
+                prompt,
+                padding="max_length",
+                max_length=self.tokenizer.model_max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+            text_input_ids = text_inputs.input_ids
+            untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+                text_input_ids, untruncated_ids
+            ):
+                removed_text = self.tokenizer.batch_decode(
+                    untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+                )
+                logger.warning(
+                    "The following part of your input was truncated because CLIP can only handle sequences up to"
+                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+                )
+
+            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+                attention_mask = text_inputs.attention_mask.to(device)
+            else:
+                attention_mask = None
+
+            prompt_embeds = self.text_encoder(
+                text_input_ids.to(device),
+                attention_mask=attention_mask,
+            )
+            prompt_embeds = prompt_embeds[0]
+
+        prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
+
+        bs_embed, seq_len, _ = prompt_embeds.shape
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+        prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+        # get unconditional embeddings for classifier free guidance
+        if do_classifier_free_guidance and negative_prompt_embeds is None:
+            uncond_tokens: List[str]
+            if negative_prompt is None:
+                uncond_tokens = [""] * batch_size
+            elif type(prompt) is not type(negative_prompt):
+                raise TypeError(
+                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+                    f" {type(prompt)}."
+                )
+            elif isinstance(negative_prompt, str):
+                uncond_tokens = [negative_prompt]
+            elif batch_size != len(negative_prompt):
+                raise ValueError(
+                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+                    " the batch size of `prompt`."
+                )
+            else:
+                uncond_tokens = negative_prompt
+
+            max_length = prompt_embeds.shape[1]
+            uncond_input = self.tokenizer(
+                uncond_tokens,
+                padding="max_length",
+                max_length=max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+
+            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+                attention_mask = uncond_input.attention_mask.to(device)
+            else:
+                attention_mask = None
+
+            negative_prompt_embeds = self.text_encoder(
+                uncond_input.input_ids.to(device),
+                attention_mask=attention_mask,
+            )
+            negative_prompt_embeds = negative_prompt_embeds[0]
+
+        if do_classifier_free_guidance:
+            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+            seq_len = negative_prompt_embeds.shape[1]
+
+            negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
+
+            negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+            # For classifier free guidance, we need to do two forward passes.
+            # Here we concatenate the unconditional and text embeddings into a single batch
+            # to avoid doing two forward passes
+            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+
+        return prompt_embeds
+
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        else:
+            has_nsfw_concept = None
+        return image, has_nsfw_concept
+
+    def decode_latents(self, latents):
+        latents = 1 / self.vae.config.scaling_factor * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    def check_inputs(
+        self,
+        prompt,
+        image,
+        mask_image,
+        controlnet_conditioning_image,
+        height,
+        width,
+        callback_steps,
+        negative_prompt=None,
+        prompt_embeds=None,
+        negative_prompt_embeds=None,
+        strength=None,
+    ):
+        if height % 8 != 0 or width % 8 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+        if (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if negative_prompt is not None and negative_prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+            )
+
+        if prompt_embeds is not None and negative_prompt_embeds is not None:
+            if prompt_embeds.shape != negative_prompt_embeds.shape:
+                raise ValueError(
+                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+                    f" {negative_prompt_embeds.shape}."
+                )
+
+        controlnet_cond_image_is_pil = isinstance(controlnet_conditioning_image, PIL.Image.Image)
+        controlnet_cond_image_is_tensor = isinstance(controlnet_conditioning_image, torch.Tensor)
+        controlnet_cond_image_is_pil_list = isinstance(controlnet_conditioning_image, list) and isinstance(
+            controlnet_conditioning_image[0], PIL.Image.Image
+        )
+        controlnet_cond_image_is_tensor_list = isinstance(controlnet_conditioning_image, list) and isinstance(
+            controlnet_conditioning_image[0], torch.Tensor
+        )
+
+        if (
+            not controlnet_cond_image_is_pil
+            and not controlnet_cond_image_is_tensor
+            and not controlnet_cond_image_is_pil_list
+            and not controlnet_cond_image_is_tensor_list
+        ):
+            raise TypeError(
+                "image must be passed and be one of PIL image, torch tensor, list of PIL images, or list of torch tensors"
+            )
+
+        if controlnet_cond_image_is_pil:
+            controlnet_cond_image_batch_size = 1
+        elif controlnet_cond_image_is_tensor:
+            controlnet_cond_image_batch_size = controlnet_conditioning_image.shape[0]
+        elif controlnet_cond_image_is_pil_list:
+            controlnet_cond_image_batch_size = len(controlnet_conditioning_image)
+        elif controlnet_cond_image_is_tensor_list:
+            controlnet_cond_image_batch_size = len(controlnet_conditioning_image)
+
+        if prompt is not None and isinstance(prompt, str):
+            prompt_batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            prompt_batch_size = len(prompt)
+        elif prompt_embeds is not None:
+            prompt_batch_size = prompt_embeds.shape[0]
+
+        if controlnet_cond_image_batch_size != 1 and controlnet_cond_image_batch_size != prompt_batch_size:
+            raise ValueError(
+                f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {controlnet_cond_image_batch_size}, prompt batch size: {prompt_batch_size}"
+            )
+
+        if isinstance(image, torch.Tensor) and not isinstance(mask_image, torch.Tensor):
+            raise TypeError("if `image` is a tensor, `mask_image` must also be a tensor")
+
+        if isinstance(image, PIL.Image.Image) and not isinstance(mask_image, PIL.Image.Image):
+            raise TypeError("if `image` is a PIL image, `mask_image` must also be a PIL image")
+
+        if isinstance(image, torch.Tensor):
+            if image.ndim != 3 and image.ndim != 4:
+                raise ValueError("`image` must have 3 or 4 dimensions")
+
+            if mask_image.ndim != 2 and mask_image.ndim != 3 and mask_image.ndim != 4:
+                raise ValueError("`mask_image` must have 2, 3, or 4 dimensions")
+
+            if image.ndim == 3:
+                image_batch_size = 1
+                image_channels, image_height, image_width = image.shape
+            elif image.ndim == 4:
+                image_batch_size, image_channels, image_height, image_width = image.shape
+
+            if mask_image.ndim == 2:
+                mask_image_batch_size = 1
+                mask_image_channels = 1
+                mask_image_height, mask_image_width = mask_image.shape
+            elif mask_image.ndim == 3:
+                mask_image_channels = 1
+                mask_image_batch_size, mask_image_height, mask_image_width = mask_image.shape
+            elif mask_image.ndim == 4:
+                mask_image_batch_size, mask_image_channels, mask_image_height, mask_image_width = mask_image.shape
+
+            if image_channels != 3:
+                raise ValueError("`image` must have 3 channels")
+
+            if mask_image_channels != 1:
+                raise ValueError("`mask_image` must have 1 channel")
+
+            if image_batch_size != mask_image_batch_size:
+                raise ValueError("`image` and `mask_image` mush have the same batch sizes")
+
+            if image_height != mask_image_height or image_width != mask_image_width:
+                raise ValueError("`image` and `mask_image` must have the same height and width dimensions")
+
+            if image.min() < -1 or image.max() > 1:
+                raise ValueError("`image` should be in range [-1, 1]")
+
+            if mask_image.min() < 0 or mask_image.max() > 1:
+                raise ValueError("`mask_image` should be in range [0, 1]")
+        else:
+            mask_image_channels = 1
+            image_channels = 3
+
+        single_image_latent_channels = self.vae.config.latent_channels
+
+        total_latent_channels = single_image_latent_channels * 2 + mask_image_channels
+
+        if total_latent_channels != self.unet.config.in_channels:
+            raise ValueError(
+                f"The config of `pipeline.unet` expects {self.unet.config.in_channels} but received"
+                f" non inpainting latent channels: {single_image_latent_channels},"
+                f" mask channels: {mask_image_channels}, and masked image channels: {single_image_latent_channels}."
+                f" Please verify the config of `pipeline.unet` and the `mask_image` and `image` inputs."
+            )
+
+        if strength < 0 or strength > 1:
+            raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
+
+    def get_timesteps(self, num_inference_steps, strength, device):
+        # get the original timestep using init_timestep
+        init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+
+        t_start = max(num_inference_steps - init_timestep, 0)
+        timesteps = self.scheduler.timesteps[t_start:]
+
+        return timesteps, num_inference_steps - t_start
+
+    def prepare_latents(self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None):
+        if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
+            raise ValueError(
+                f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
+            )
+
+        image = image.to(device=device, dtype=dtype)
+
+        batch_size = batch_size * num_images_per_prompt
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        if isinstance(generator, list):
+            init_latents = [
+                self.vae.encode(image[i : i + 1]).latent_dist.sample(generator[i]) for i in range(batch_size)
+            ]
+            init_latents = torch.cat(init_latents, dim=0)
+        else:
+            init_latents = self.vae.encode(image).latent_dist.sample(generator)
+
+        init_latents = self.vae.config.scaling_factor * init_latents
+
+        if batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] == 0:
+            raise ValueError(
+                f"Cannot duplicate `image` of batch size {init_latents.shape[0]} to {batch_size} text prompts."
+            )
+        else:
+            init_latents = torch.cat([init_latents], dim=0)
+
+        shape = init_latents.shape
+        noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+
+        # get latents
+        init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
+        latents = init_latents
+
+        return latents
+
+    def prepare_mask_latents(self, mask_image, batch_size, height, width, dtype, device, do_classifier_free_guidance):
+        # resize the mask to latents shape as we concatenate the mask to the latents
+        # we do that before converting to dtype to avoid breaking in case we're using cpu_offload
+        # and half precision
+        mask_image = F.interpolate(mask_image, size=(height // self.vae_scale_factor, width // self.vae_scale_factor))
+        mask_image = mask_image.to(device=device, dtype=dtype)
+
+        # duplicate mask for each generation per prompt, using mps friendly method
+        if mask_image.shape[0] < batch_size:
+            if not batch_size % mask_image.shape[0] == 0:
+                raise ValueError(
+                    "The passed mask and the required batch size don't match. Masks are supposed to be duplicated to"
+                    f" a total batch size of {batch_size}, but {mask_image.shape[0]} masks were passed. Make sure the number"
+                    " of masks that you pass is divisible by the total requested batch size."
+                )
+            mask_image = mask_image.repeat(batch_size // mask_image.shape[0], 1, 1, 1)
+
+        mask_image = torch.cat([mask_image] * 2) if do_classifier_free_guidance else mask_image
+
+        mask_image_latents = mask_image
+
+        return mask_image_latents
+
+    def prepare_masked_image_latents(
+        self, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance
+    ):
+        masked_image = masked_image.to(device=device, dtype=dtype)
+
+        # encode the mask image into latents space so we can concatenate it to the latents
+        if isinstance(generator, list):
+            masked_image_latents = [
+                self.vae.encode(masked_image[i : i + 1]).latent_dist.sample(generator=generator[i])
+                for i in range(batch_size)
+            ]
+            masked_image_latents = torch.cat(masked_image_latents, dim=0)
+        else:
+            masked_image_latents = self.vae.encode(masked_image).latent_dist.sample(generator=generator)
+        masked_image_latents = self.vae.config.scaling_factor * masked_image_latents
+
+        # duplicate masked_image_latents for each generation per prompt, using mps friendly method
+        if masked_image_latents.shape[0] < batch_size:
+            if not batch_size % masked_image_latents.shape[0] == 0:
+                raise ValueError(
+                    "The passed images and the required batch size don't match. Images are supposed to be duplicated"
+                    f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed."
+                    " Make sure the number of images that you pass is divisible by the total requested batch size."
+                )
+            masked_image_latents = masked_image_latents.repeat(batch_size // masked_image_latents.shape[0], 1, 1, 1)
+
+        masked_image_latents = (
+            torch.cat([masked_image_latents] * 2) if do_classifier_free_guidance else masked_image_latents
+        )
+
+        # aligning device to prevent device errors when concating it with the latent model input
+        masked_image_latents = masked_image_latents.to(device=device, dtype=dtype)
+        return masked_image_latents
+
+    def _default_height_width(self, height, width, image):
+        if isinstance(image, list):
+            image = image[0]
+
+        if height is None:
+            if isinstance(image, PIL.Image.Image):
+                height = image.height
+            elif isinstance(image, torch.Tensor):
+                height = image.shape[3]
+
+            height = (height // 8) * 8  # round down to nearest multiple of 8
+
+        if width is None:
+            if isinstance(image, PIL.Image.Image):
+                width = image.width
+            elif isinstance(image, torch.Tensor):
+                width = image.shape[2]
+
+            width = (width // 8) * 8  # round down to nearest multiple of 8
+
+        return height, width
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        image: Union[torch.Tensor, PIL.Image.Image] = None,
+        mask_image: Union[torch.Tensor, PIL.Image.Image] = None,
+        controlnet_conditioning_image: Union[
+            torch.FloatTensor, PIL.Image.Image, List[torch.FloatTensor], List[PIL.Image.Image]
+        ] = None,
+        strength: float = 0.8,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 50,
+        guidance_scale: float = 7.5,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        num_images_per_prompt: Optional[int] = 1,
+        eta: float = 0.0,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: int = 1,
+        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        controlnet_conditioning_scale: float = 1.0,
+        controlnet_conditioning_scale_decay: float = 0.95,
+        controlnet_steps: int = 10, 
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            image (`torch.Tensor` or `PIL.Image.Image`):
+                `Image`, or tensor representing an image batch which will be inpainted, *i.e.* parts of the image will
+                be masked out with `mask_image` and repainted according to `prompt`.
+            mask_image (`torch.Tensor` or `PIL.Image.Image`):
+                `Image`, or tensor representing an image batch, to mask `image`. White pixels in the mask will be
+                repainted, while black pixels will be preserved. If `mask_image` is a PIL image, it will be converted
+                to a single channel (luminance) before use. If it's a tensor, it should contain one color channel (L)
+                instead of 3, so the expected shape would be `(B, H, W, 1)`.
+            controlnet_conditioning_image (`torch.FloatTensor`, `PIL.Image.Image`, `List[torch.FloatTensor]` or `List[PIL.Image.Image]`):
+                The ControlNet input condition. ControlNet uses this input condition to generate guidance to Unet. If
+                the type is specified as `Torch.FloatTensor`, it is passed to ControlNet as is. PIL.Image.Image` can
+                also be accepted as an image. The control image is automatically resized to fit the output image.
+            strength (`float`, *optional*):
+                Conceptually, indicates how much to transform the reference `image`. Must be between 0 and 1. `image`
+                will be used as a starting point, adding more noise to it the larger the `strength`. The number of
+                denoising steps depends on the amount of noise initially added. When `strength` is 1, added noise will
+                be maximum and the denoising process will run for the full number of iterations specified in
+                `num_inference_steps`. A value of 1, therefore, essentially ignores `image`.
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The width in pixels of the generated image.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.
+                Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+                [`schedulers.DDIMScheduler`], will be ignored for others.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+                to make generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            callback (`Callable`, *optional*):
+                A function that will be called every `callback_steps` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+            cross_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.cross_attention](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/cross_attention.py).
+            controlnet_conditioning_scale (`float`, *optional*, defaults to 1.0):
+                The outputs of the controlnet are multiplied by `controlnet_conditioning_scale` before they are added
+                to the residual in the original unet.
+        Examples:
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
+            When returning a tuple, the first element is a list with the generated images, and the second element is a
+            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
+            (nsfw) content, according to the `safety_checker`.
+        """
+        # 0. Default height and width to unet
+        height, width = self._default_height_width(height, width, controlnet_conditioning_image)
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            image,
+            mask_image,
+            controlnet_conditioning_image,
+            height,
+            width,
+            callback_steps,
+            negative_prompt,
+            prompt_embeds,
+            negative_prompt_embeds,
+            strength,
+        )
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        device = self._execution_device
+        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # corresponds to doing no classifier free guidance.
+        do_classifier_free_guidance = guidance_scale > 1.0
+
+        # 3. Encode input prompt
+        prompt_embeds = self._encode_prompt(
+            prompt,
+            device,
+            num_images_per_prompt,
+            do_classifier_free_guidance,
+            negative_prompt,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+        )
+
+        # 4. Prepare mask, image, and controlnet_conditioning_image
+        image = prepare_image(image)
+
+        mask_image = prepare_mask_image(mask_image)
+
+        controlnet_conditioning_image = prepare_controlnet_conditioning_image(
+            controlnet_conditioning_image,
+            width,
+            height,
+            batch_size * num_images_per_prompt,
+            num_images_per_prompt,
+            device,
+            self.controlnet.dtype,
+        )
+
+        masked_image = image * (mask_image < 0.5)
+
+        # 5. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
+        latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+
+        # 6. Prepare latent variables
+        latents = self.prepare_latents(
+            image,
+            latent_timestep,
+            batch_size,
+            num_images_per_prompt,
+            prompt_embeds.dtype,
+            device,
+            generator,
+        )
+
+        mask_image_latents = self.prepare_mask_latents(
+            mask_image,
+            batch_size * num_images_per_prompt,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            do_classifier_free_guidance,
+        )
+
+        masked_image_latents = self.prepare_masked_image_latents(
+            masked_image,
+            batch_size * num_images_per_prompt,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            do_classifier_free_guidance,
+        )
+
+        if do_classifier_free_guidance:
+            controlnet_conditioning_image = torch.cat([controlnet_conditioning_image] * 2)
+
+        # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+        # 8. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                # expand the latents if we are doing classifier free guidance
+                non_inpainting_latent_model_input = (
+                    torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+                )
+
+                non_inpainting_latent_model_input = self.scheduler.scale_model_input(
+                    non_inpainting_latent_model_input, t
+                )
+
+                inpainting_latent_model_input = torch.cat(
+                    [non_inpainting_latent_model_input, mask_image_latents, masked_image_latents], dim=1
+                )
+
+                down_block_res_samples, mid_block_res_sample = self.controlnet(
+                    non_inpainting_latent_model_input,
+                    t,
+                    encoder_hidden_states=prompt_embeds,
+                    controlnet_cond=controlnet_conditioning_image,
+                    return_dict=False,
+                )
+                if i <= controlnet_steps:
+                    conditioning_scale = (controlnet_conditioning_scale * controlnet_conditioning_scale_decay ** i)
+                else:
+                    conditioning_scale = 0.0
+
+                down_block_res_samples = [
+                    down_block_res_sample * conditioning_scale
+                    for down_block_res_sample in down_block_res_samples
+                ]
+                mid_block_res_sample *= conditioning_scale
+
+                # predict the noise residual
+                noise_pred = self.unet(
+                    inpainting_latent_model_input,
+                    t,
+                    encoder_hidden_states=prompt_embeds,
+                    cross_attention_kwargs=cross_attention_kwargs,
+                    down_block_additional_residuals=down_block_res_samples,
+                    mid_block_additional_residual=mid_block_res_sample,
+                ).sample
+
+                # perform guidance
+                if do_classifier_free_guidance:
+                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+                    if callback is not None and i % callback_steps == 0:
+                        callback(i, t, latents)
+
+        # If we do sequential model offloading, let's offload unet and controlnet
+        # manually for max memory savings
+        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+            self.unet.to("cpu")
+            self.controlnet.to("cpu")
+            torch.cuda.empty_cache()
+
+        if output_type == "latent":
+            image = latents
+            has_nsfw_concept = None
+        elif output_type == "pil":
+            # 8. Post-processing
+            image = self.decode_latents(latents)
+
+            # 9. Run safety checker
+            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+
+            # 10. Convert to PIL
+            image = self.numpy_to_pil(image)
+        else:
+            # 8. Post-processing
+            image = self.decode_latents(latents)
+
+            # 9. Run safety checker
+            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+
+        # Offload last model to CPU
+        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+            self.final_offload_hook.offload()
+
+        if not return_dict:
+            return (image, has_nsfw_concept)
+
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)