File size: 7,802 Bytes
cc579f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
#!/usr/bin/env python

import cv2
import numpy as np
import torch
import random
import base64
import json
import threading
import uuid
import math

import io
from PIL import Image

from diffusers import AutoencoderKL, StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler,StableDiffusionControlNetImg2ImgPipeline,StableDiffusionXLControlNetPipeline,DiffusionPipeline
from diffusers.utils import load_image
from transformers import pipeline

import gradio as gr

vae = AutoencoderKL.from_pretrained("stabilityai/sd-vae-ft-mse", torch_dtype=torch.float16)


canny_controlnet = ControlNetModel.from_pretrained("lllyasviel/control_v11p_sd15_canny", torch_dtype=torch.float16)
canny_pipe = StableDiffusionControlNetPipeline.from_pretrained(
    "SG161222/Realistic_Vision_V3.0_VAE", controlnet=canny_controlnet, torch_dtype=torch.float16, use_safetensors=True
)

canny_controlnet_tile = ControlNetModel.from_pretrained("lllyasviel/control_v11f1e_sd15_tile", torch_dtype=torch.float16)
canny_pipe_img2img = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
    "SG161222/Realistic_Vision_V3.0_VAE", controlnet=canny_controlnet_tile, torch_dtype=torch.float16, use_safetensors=True
)
canny_pipe_img2img.enable_model_cpu_offload()
canny_pipe_img2img.enable_xformers_memory_efficient_attention()


canny_pipe.scheduler = UniPCMultistepScheduler.from_config(canny_pipe.scheduler.config)
canny_pipe.enable_model_cpu_offload()
canny_pipe.enable_xformers_memory_efficient_attention()

controlnet_xl = ControlNetModel.from_pretrained(
    "diffusers/controlnet-canny-sdxl-1.0",
    torch_dtype=torch.float16
)
vae_xl = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
pipe_xl = StableDiffusionXLControlNetPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    controlnet=controlnet_xl,
    vae=vae_xl,
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16",
)
pipe_xl.scheduler = UniPCMultistepScheduler.from_config(pipe_xl.scheduler.config)
pipe_xl.enable_xformers_memory_efficient_attention()
pipe_xl.enable_model_cpu_offload()

refiner = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-refiner-1.0",
    text_encoder_2=pipe_xl.text_encoder_2,
    vae=pipe_xl.vae,
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16",
)
refiner.enable_xformers_memory_efficient_attention()
refiner.enable_model_cpu_offload()

def resize_image_output(im, width, height):
    im = np.array(im)   
    newSize = (width,height)
    img = cv2.resize(im, newSize, interpolation=cv2.INTER_CUBIC)
    img = Image.fromarray(img)
    return img

def resize_image(im, max_size = 590000):
    [x,y,z] = im.shape
    new_size = [0,0]


    min_size = 262144
    if x*y > max_size:
        scale_ratio = math.sqrt((x*y)/max_size)
        new_size[0] = int(x / scale_ratio)
        new_size[1] = int(y / scale_ratio)
    elif x*y <= min_size:
        scale_ratio = math.sqrt((x*y)/min_size)
        new_size[0] = int(x / scale_ratio)
        new_size[1] = int(y / scale_ratio)
    else:
        new_size[0] = int(x)
        new_size[1] = int(y)
        
    height = (new_size[0] // 8) * 8
    width = (new_size[1] // 8) * 8
    
    newSize = (width,height)
    img = cv2.resize(im, newSize, interpolation=cv2.INTER_CUBIC)
    return img

def process_canny_tile(input_image,control_image, x ,y, prompt, a_prompt, n_prompt, num_samples, image_resolution, ddim_steps, guess_mode, strength_conditioning, scale, seed, eta, low_threshold, high_threshold):

    image = input_image

    return canny_pipe_img2img(
        prompt = '',
        image=image,
        control_image = image,
        num_inference_steps=20,
        guidance_scale=4,
        strength = 0.3,
        guess_mode = True,
        negative_prompt=n_prompt,
        num_images_per_prompt=1,
        eta=eta,
        generator=torch.Generator(device="cpu").manual_seed(seed)
    )

def process_canny(input_image,x ,y, prompt, a_prompt, n_prompt, num_samples, image_resolution, ddim_steps, guess_mode, strength, scale, seed, eta, low_threshold, high_threshold):

    image = input_image

    print(strength)


    return canny_pipe(
        prompt=','.join([prompt,a_prompt]),
        image=image,
        height=x,
        width=y,
        num_inference_steps=ddim_steps,
        guidance_scale=scale,
        negative_prompt=n_prompt,
        num_images_per_prompt=num_samples,
        eta=eta,
        controlnet_conditioning_scale=strength,
        generator=torch.Generator(device="cpu").manual_seed(seed)
    )

def process_canny_sdxl(input_image,x ,y, prompt, a_prompt, n_prompt, num_samples, image_resolution, ddim_steps, guess_mode, strength, scale, seed, eta, low_threshold, high_threshold):

    image = input_image
    
    image = pipe_xl(
        prompt=','.join([prompt,a_prompt]),
        image=image,
        height=x,
        width=y,
        num_inference_steps=ddim_steps,
        guidance_scale=scale,
        negative_prompt=n_prompt,
        num_images_per_prompt=num_samples,
        eta=eta,
        controlnet_conditioning_scale=strength,
        generator=torch.Generator(device="cpu").manual_seed(seed),
        output_type="latent"
    ).images
    
    return refiner(
    prompt=prompt,
    num_inference_steps=ddim_steps,
    num_images_per_prompt=num_samples,
    denoising_start=0.8,
    image=image,
    )


def process(image, prompt, a_prompt, n_prompt, ddim_steps, strength, scale, seed, eta, low_threshold, high_threshold):
    image = load_image(image)
    image = np.array(image)
    [x_orig,y_orig,z_orig] = image.shape
    image = resize_image(image)
    [x,y,z] = image.shape

    image = cv2.Canny(image, low_threshold, high_threshold)
    image = image[:, :, None]
    image = np.concatenate([image, image, image], axis=2)
    image = Image.fromarray(image)

    return process_canny(image,x,y, prompt, a_prompt, n_prompt, 1, None, ddim_steps, False, float(strength), scale, seed, eta, low_threshold, high_threshold)


demo = gr.Blocks().queue()

with demo:
    with gr.Row():
        gr.Markdown("## Control Stable Diffusion with Canny Edge Maps")

    with gr.Row():
        with gr.Column():
            input_image = gr.Image(type="pil", label="Input Image")
            input_prompt = gr.Textbox()
            run_button = gr.Button(label="Run")

            with gr.Accordion("Advanced Options"):
                strength = gr.Slider(label="Control Strength", minimum=0.0, maximum=2.0, value=1.0, step=0.01)
                low_threshold = gr.Slider(label="Canny low threshold", minimum=1, maximum=255, value=100, step=1)
                high_threshold = gr.Slider(label="Canny high threshold", minimum=1, maximum=255, value=200, step=1)
                ddim_steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=20, step=1)
                scale = gr.Slider(label="Guidance Scale", minimum=0.1, maximum=30.0, value=7.5, step=0.1)  # default value was 9.0
                seed = gr.Slider(label="Seed", minimum=-1, maximum=2147483647, step=1, randomize=True)
                eta = gr.Number(label="eta (DDIM)", value=0.0)
                a_prompt = gr.Textbox(label="Added Prompt", value='best quality, extremely detailed')
                n_prompt = gr.Textbox(label="Negative Prompt",
                                      value='longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality')

        with gr.Column():
            result = gr.outputs.Image(label='Output', type="pil")

    ips = [input_image, input_prompt, a_prompt, n_prompt, ddim_steps, strength, scale, seed, eta, low_threshold, high_threshold]
    run_button.click(fn=process, inputs=ips, outputs=[result])


demo.launch()