Upload 3 files
Browse files- ArmorSuit_v1.safetensors +3 -0
- Cyberspace_background_composer.safetensors +3 -0
- app.py +186 -4
ArmorSuit_v1.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:a16b4cf713537c007b8cf1dc44e13e2730503b3a822760cbf46f9d7cb8fb5a0b
|
3 |
+
size 151112956
|
Cyberspace_background_composer.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:5a4cc2f5cb421a4859f96a8e228d37cad7ef4854c87e9602ccb83cbe435652d5
|
3 |
+
size 75613462
|
app.py
CHANGED
@@ -1,7 +1,189 @@
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
|
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
import gradio as gr
|
3 |
+
from gradio import processing_utils, utils
|
4 |
+
from PIL import Image
|
5 |
+
import random
|
6 |
+
from diffusers import (
|
7 |
+
DiffusionPipeline,
|
8 |
+
AutoencoderKL,
|
9 |
+
StableDiffusionControlNetPipeline,
|
10 |
+
ControlNetModel,
|
11 |
+
StableDiffusionLatentUpscalePipeline,
|
12 |
+
StableDiffusionImg2ImgPipeline,
|
13 |
+
StableDiffusionControlNetImg2ImgPipeline,
|
14 |
+
DPMSolverMultistepScheduler, # <-- Added import
|
15 |
+
EulerDiscreteScheduler # <-- Added import
|
16 |
+
)
|
17 |
|
18 |
+
import time
|
19 |
+
from share_btn import community_icon_html, loading_icon_html, share_js
|
20 |
+
import user_history
|
21 |
+
from illusion_style import css
|
22 |
|
23 |
+
BASE_MODEL = "SG161222/Realistic_Vision_V5.1_noVAE"
|
24 |
+
|
25 |
+
# Initialize both pipelines
|
26 |
+
vae = AutoencoderKL.from_pretrained("stabilityai/sd-vae-ft-mse", torch_dtype=torch.float16)
|
27 |
+
#init_pipe = DiffusionPipeline.from_pretrained("SG161222/Realistic_Vision_V5.1_noVAE", torch_dtype=torch.float16)
|
28 |
+
controlnet = ControlNetModel.from_pretrained("monster-labs/control_v1p_sd15_qrcode_monster", torch_dtype=torch.float16)#, torch_dtype=torch.float16)
|
29 |
+
main_pipe = StableDiffusionControlNetPipeline.from_pretrained(
|
30 |
+
BASE_MODEL,
|
31 |
+
controlnet=controlnet,
|
32 |
+
vae=vae,
|
33 |
+
safety_checker=None,
|
34 |
+
torch_dtype=torch.float16,
|
35 |
+
).to("cuda")
|
36 |
+
|
37 |
+
#main_pipe.unet = torch.compile(main_pipe.unet, mode="reduce-overhead", fullgraph=True)
|
38 |
+
#main_pipe.unet.to(memory_format=torch.channels_last)
|
39 |
+
#main_pipe.unet = torch.compile(main_pipe.unet, mode="reduce-overhead", fullgraph=True)
|
40 |
+
#model_id = "stabilityai/sd-x2-latent-upscaler"
|
41 |
+
image_pipe = StableDiffusionControlNetImg2ImgPipeline(**main_pipe.components)
|
42 |
+
|
43 |
+
#image_pipe.unet = torch.compile(image_pipe.unet, mode="reduce-overhead", fullgraph=True)
|
44 |
+
#upscaler = StableDiffusionLatentUpscalePipeline.from_pretrained(model_id, torch_dtype=torch.float16)
|
45 |
+
#upscaler.to("cuda")
|
46 |
+
|
47 |
+
|
48 |
+
# Sampler map
|
49 |
+
SAMPLER_MAP = {
|
50 |
+
"DPM++ Karras SDE": lambda config: DPMSolverMultistepScheduler.from_config(config, use_karras=True, algorithm_type="sde-dpmsolver++"),
|
51 |
+
"Euler": lambda config: EulerDiscreteScheduler.from_config(config),
|
52 |
+
}
|
53 |
+
|
54 |
+
def center_crop_resize(img, output_size=(512, 512)):
|
55 |
+
width, height = img.size
|
56 |
+
|
57 |
+
# Calculate dimensions to crop to the center
|
58 |
+
new_dimension = min(width, height)
|
59 |
+
left = (width - new_dimension)/2
|
60 |
+
top = (height - new_dimension)/2
|
61 |
+
right = (width + new_dimension)/2
|
62 |
+
bottom = (height + new_dimension)/2
|
63 |
+
|
64 |
+
# Crop and resize
|
65 |
+
img = img.crop((left, top, right, bottom))
|
66 |
+
img = img.resize(output_size)
|
67 |
+
|
68 |
+
return img
|
69 |
+
|
70 |
+
def common_upscale(samples, width, height, upscale_method, crop=False):
|
71 |
+
if crop == "center":
|
72 |
+
old_width = samples.shape[3]
|
73 |
+
old_height = samples.shape[2]
|
74 |
+
old_aspect = old_width / old_height
|
75 |
+
new_aspect = width / height
|
76 |
+
x = 0
|
77 |
+
y = 0
|
78 |
+
if old_aspect > new_aspect:
|
79 |
+
x = round((old_width - old_width * (new_aspect / old_aspect)) / 2)
|
80 |
+
elif old_aspect < new_aspect:
|
81 |
+
y = round((old_height - old_height * (old_aspect / new_aspect)) / 2)
|
82 |
+
s = samples[:,:,y:old_height-y,x:old_width-x]
|
83 |
+
else:
|
84 |
+
s = samples
|
85 |
+
|
86 |
+
return torch.nn.functional.interpolate(s, size=(height, width), mode=upscale_method)
|
87 |
+
|
88 |
+
def upscale(samples, upscale_method, scale_by):
|
89 |
+
#s = samples.copy()
|
90 |
+
width = round(samples["images"].shape[3] * scale_by)
|
91 |
+
height = round(samples["images"].shape[2] * scale_by)
|
92 |
+
s = common_upscale(samples["images"], width, height, upscale_method, "disabled")
|
93 |
+
return (s)
|
94 |
+
|
95 |
+
def check_inputs(prompt: str, control_image: Image.Image):
|
96 |
+
if control_image is None:
|
97 |
+
raise gr.Error("Please select or upload an Input Illusion")
|
98 |
+
if prompt is None or prompt == "":
|
99 |
+
raise gr.Error("Prompt is required")
|
100 |
+
|
101 |
+
def convert_to_pil(base64_image):
|
102 |
+
pil_image = processing_utils.decode_base64_to_image(base64_image)
|
103 |
+
return pil_image
|
104 |
+
|
105 |
+
def convert_to_base64(pil_image):
|
106 |
+
base64_image = processing_utils.encode_pil_to_base64(pil_image)
|
107 |
+
return base64_image
|
108 |
+
|
109 |
+
# Inference function
|
110 |
+
def inference(
|
111 |
+
control_image: Image.Image,
|
112 |
+
prompt: str,
|
113 |
+
negative_prompt: str,
|
114 |
+
guidance_scale: float = 8.0,
|
115 |
+
controlnet_conditioning_scale: float = 1,
|
116 |
+
control_guidance_start: float = 1,
|
117 |
+
control_guidance_end: float = 1,
|
118 |
+
upscaler_strength: float = 0.5,
|
119 |
+
seed: int = -1,
|
120 |
+
sampler = "DPM++ Karras SDE",
|
121 |
+
progress = gr.Progress(track_tqdm=True),
|
122 |
+
profile: gr.OAuthProfile | None = None,
|
123 |
+
):
|
124 |
+
start_time = time.time()
|
125 |
+
start_time_struct = time.localtime(start_time)
|
126 |
+
start_time_formatted = time.strftime("%H:%M:%S", start_time_struct)
|
127 |
+
print(f"Inference started at {start_time_formatted}")
|
128 |
+
|
129 |
+
# Generate the initial image
|
130 |
+
#init_image = init_pipe(prompt).images[0]
|
131 |
+
|
132 |
+
# Rest of your existing code
|
133 |
+
control_image_small = center_crop_resize(control_image)
|
134 |
+
control_image_large = center_crop_resize(control_image, (1024, 1024))
|
135 |
+
|
136 |
+
main_pipe.scheduler = SAMPLER_MAP[sampler](main_pipe.scheduler.config)
|
137 |
+
my_seed = random.randint(0, 2**32 - 1) if seed == -1 else seed
|
138 |
+
generator = torch.Generator(device="cuda").manual_seed(my_seed)
|
139 |
+
|
140 |
+
out = main_pipe(
|
141 |
+
prompt=prompt,
|
142 |
+
negative_prompt=negative_prompt,
|
143 |
+
image=control_image_small,
|
144 |
+
guidance_scale=float(guidance_scale),
|
145 |
+
controlnet_conditioning_scale=float(controlnet_conditioning_scale),
|
146 |
+
generator=generator,
|
147 |
+
control_guidance_start=float(control_guidance_start),
|
148 |
+
control_guidance_end=float(control_guidance_end),
|
149 |
+
num_inference_steps=15,
|
150 |
+
output_type="latent"
|
151 |
+
)
|
152 |
+
upscaled_latents = upscale(out, "nearest-exact", 2)
|
153 |
+
out_image = image_pipe(
|
154 |
+
prompt=prompt,
|
155 |
+
negative_prompt=negative_prompt,
|
156 |
+
control_image=control_image_large,
|
157 |
+
image=upscaled_latents,
|
158 |
+
guidance_scale=float(guidance_scale),
|
159 |
+
generator=generator,
|
160 |
+
num_inference_steps=20,
|
161 |
+
strength=upscaler_strength,
|
162 |
+
control_guidance_start=float(control_guidance_start),
|
163 |
+
control_guidance_end=float(control_guidance_end),
|
164 |
+
controlnet_conditioning_scale=float(controlnet_conditioning_scale)
|
165 |
+
)
|
166 |
+
end_time = time.time()
|
167 |
+
end_time_struct = time.localtime(end_time)
|
168 |
+
end_time_formatted = time.strftime("%H:%M:%S", end_time_struct)
|
169 |
+
print(f"Inference ended at {end_time_formatted}, taking {end_time-start_time}s")
|
170 |
+
|
171 |
+
# Save image + metadata
|
172 |
+
user_history.save_image(
|
173 |
+
label=prompt,
|
174 |
+
image=out_image["images"][0],
|
175 |
+
profile=profile,
|
176 |
+
metadata={
|
177 |
+
"prompt": prompt,
|
178 |
+
"negative_prompt": negative_prompt,
|
179 |
+
"guidance_scale": guidance_scale,
|
180 |
+
"controlnet_conditioning_scale": controlnet_conditioning_scale,
|
181 |
+
"control_guidance_start": control_guidance_start,
|
182 |
+
"control_guidance_end": control_guidance_end,
|
183 |
+
"upscaler_strength": upscaler_strength,
|
184 |
+
"seed": seed,
|
185 |
+
"sampler": sampler,
|
186 |
+
},
|
187 |
+
)
|
188 |
+
|
189 |
+
return out_image["images"][0], gr.update(visible=True), gr.update(visible=True), my_seed
|