Commit
•
431f639
1
Parent(s):
7dfd19e
Update app.py
Browse files
app.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
# Thank AK. https://huggingface.co/spaces/akhaliq/cool-japan-diffusion-2-1-0/blob/main/app.py
|
2 |
-
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline, EulerAncestralDiscreteScheduler
|
3 |
from transformers import CLIPFeatureExtractor
|
4 |
import gradio as gr
|
5 |
import torch
|
@@ -27,6 +27,10 @@ pipe_i2i = StableDiffusionImg2ImgPipeline.from_pretrained(
|
|
27 |
)
|
28 |
pipe_i2i.enable_xformers_memory_efficient_attention()
|
29 |
|
|
|
|
|
|
|
|
|
30 |
if torch.cuda.is_available():
|
31 |
pipe = pipe.to("cuda")
|
32 |
pipe_i2i = pipe_i2i.to("cuda")
|
@@ -45,27 +49,36 @@ def inference(prompt, guidance, steps, image_size="Square", seed=0, img=None, st
|
|
45 |
if(image_size=="Portrait"):
|
46 |
height=1024
|
47 |
width=768
|
|
|
48 |
#pipe.enable_attention_slicing()
|
49 |
elif(image_size=="Landscape"):
|
50 |
height=768
|
51 |
width=1024
|
|
|
52 |
#pipe.enable_attention_slicing()
|
53 |
elif(image_size=="Highreso."):
|
54 |
height=1024
|
55 |
-
width=1024
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
#pipe.enable_attention_slicing()
|
57 |
else:
|
58 |
height=768
|
59 |
width=768
|
|
|
60 |
#pipe.enable_attention_slicing()
|
61 |
|
62 |
print(prompt,neg_prompt)
|
63 |
|
64 |
try:
|
65 |
if img is not None:
|
66 |
-
return img_to_img(prompt, neg_prompt, img, strength, guidance, steps, width, height, generator), None
|
67 |
else:
|
68 |
-
return txt_to_img(prompt, neg_prompt, guidance, steps, width, height, generator), None
|
69 |
except Exception as e:
|
70 |
return None, error_str(e)
|
71 |
def auto_prompt_correction(prompt_ui,neg_prompt_ui,cool_japan_type_ui,disable_auto_prompt_correction):
|
@@ -123,22 +136,65 @@ def auto_prompt_correction(prompt_ui,neg_prompt_ui,cool_japan_type_ui,disable_au
|
|
123 |
|
124 |
return prompt,neg_prompt
|
125 |
|
126 |
-
def txt_to_img(prompt, neg_prompt, guidance, steps, width, height, generator):
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
135 |
|
136 |
return result.images[0]
|
137 |
|
138 |
-
def img_to_img(prompt, neg_prompt, img, strength, guidance, steps, width, height, generator):
|
139 |
ratio = min(height / img.height, width / img.width)
|
140 |
img = img.resize((int(img.width * ratio), int(img.height * ratio)), Image.LANCZOS)
|
141 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
142 |
prompt,
|
143 |
negative_prompt = neg_prompt,
|
144 |
init_image = img,
|
@@ -148,6 +204,7 @@ def img_to_img(prompt, neg_prompt, img, strength, guidance, steps, width, height
|
|
148 |
#width = width,
|
149 |
#height = height,
|
150 |
generator = generator)
|
|
|
151 |
|
152 |
return result.images[0]
|
153 |
|
@@ -199,7 +256,7 @@ with gr.Blocks(css=css) as demo:
|
|
199 |
neg_prompt = gr.Textbox(label="Negative prompt", placeholder="What to exclude from the image")
|
200 |
disable_auto_prompt_correction = gr.Checkbox(label="Disable auto prompt corretion.")
|
201 |
with gr.Row():
|
202 |
-
image_size=gr.Radio(["Portrait","Landscape","Square","Highreso."])
|
203 |
image_size.show_label=False
|
204 |
image_size.value="Square"
|
205 |
|
|
|
1 |
# Thank AK. https://huggingface.co/spaces/akhaliq/cool-japan-diffusion-2-1-0/blob/main/app.py
|
2 |
+
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline, EulerAncestralDiscreteScheduler,StableDiffusionLatentUpscalePipeline
|
3 |
from transformers import CLIPFeatureExtractor
|
4 |
import gradio as gr
|
5 |
import torch
|
|
|
27 |
)
|
28 |
pipe_i2i.enable_xformers_memory_efficient_attention()
|
29 |
|
30 |
+
upscaler = StableDiffusionLatentUpscalePipeline.from_pretrained("stabilityai/sd-x2-latent-upscaler", torch_dtype=torch.float16)
|
31 |
+
upscaler.enable_xformers_memory_efficient_attention()
|
32 |
+
upscaler.to("cuda")
|
33 |
+
|
34 |
if torch.cuda.is_available():
|
35 |
pipe = pipe.to("cuda")
|
36 |
pipe_i2i = pipe_i2i.to("cuda")
|
|
|
49 |
if(image_size=="Portrait"):
|
50 |
height=1024
|
51 |
width=768
|
52 |
+
superreso=False
|
53 |
#pipe.enable_attention_slicing()
|
54 |
elif(image_size=="Landscape"):
|
55 |
height=768
|
56 |
width=1024
|
57 |
+
superreso=False
|
58 |
#pipe.enable_attention_slicing()
|
59 |
elif(image_size=="Highreso."):
|
60 |
height=1024
|
61 |
+
width=1024
|
62 |
+
superreso=False
|
63 |
+
#pipe.enable_attention_slicing()
|
64 |
+
elif(image_size=="Superreso."):
|
65 |
+
height=1024
|
66 |
+
width=1024
|
67 |
+
superreso=True
|
68 |
#pipe.enable_attention_slicing()
|
69 |
else:
|
70 |
height=768
|
71 |
width=768
|
72 |
+
superreso=False
|
73 |
#pipe.enable_attention_slicing()
|
74 |
|
75 |
print(prompt,neg_prompt)
|
76 |
|
77 |
try:
|
78 |
if img is not None:
|
79 |
+
return img_to_img(prompt, neg_prompt, img, strength, guidance, steps, width, height, generator,superreso), None
|
80 |
else:
|
81 |
+
return txt_to_img(prompt, neg_prompt, guidance, steps, width, height, generator,superreso), None
|
82 |
except Exception as e:
|
83 |
return None, error_str(e)
|
84 |
def auto_prompt_correction(prompt_ui,neg_prompt_ui,cool_japan_type_ui,disable_auto_prompt_correction):
|
|
|
136 |
|
137 |
return prompt,neg_prompt
|
138 |
|
139 |
+
def txt_to_img(prompt, neg_prompt, guidance, steps, width, height, generator,superreso=False):
|
140 |
+
if(superreso):
|
141 |
+
low_res_latents = pipe(
|
142 |
+
prompt,
|
143 |
+
negative_prompt = neg_prompt,
|
144 |
+
num_inference_steps = int(steps),
|
145 |
+
guidance_scale = guidance,
|
146 |
+
width = width,
|
147 |
+
height = height,
|
148 |
+
output_type="latent",
|
149 |
+
generator = generator)
|
150 |
+
low_res_latents = pipeline(prompt, generator=generator, output_type="latent").images
|
151 |
+
|
152 |
+
result = upscaler(
|
153 |
+
prompt=prompt,
|
154 |
+
negative_prompt = neg_prompt,
|
155 |
+
image=low_res_latents,
|
156 |
+
num_inference_steps=20,
|
157 |
+
guidance_scale=guidance,
|
158 |
+
generator=generator,
|
159 |
+
)
|
160 |
+
else:
|
161 |
+
result = pipe(
|
162 |
+
prompt,
|
163 |
+
negative_prompt = neg_prompt,
|
164 |
+
num_inference_steps = int(steps),
|
165 |
+
guidance_scale = guidance,
|
166 |
+
width = width,
|
167 |
+
height = height,
|
168 |
+
generator = generator)
|
169 |
|
170 |
return result.images[0]
|
171 |
|
172 |
+
def img_to_img(prompt, neg_prompt, img, strength, guidance, steps, width, height, generator,superreso=False):
|
173 |
ratio = min(height / img.height, width / img.width)
|
174 |
img = img.resize((int(img.width * ratio), int(img.height * ratio)), Image.LANCZOS)
|
175 |
+
if(superreso):
|
176 |
+
low_res_latents = pipe_i2i(
|
177 |
+
prompt,
|
178 |
+
negative_prompt = neg_prompt,
|
179 |
+
init_image = img,
|
180 |
+
num_inference_steps = int(steps),
|
181 |
+
strength = strength,
|
182 |
+
guidance_scale = guidance,
|
183 |
+
#width = width,
|
184 |
+
#height = height,
|
185 |
+
output_type="latent",
|
186 |
+
generator = generator)
|
187 |
+
|
188 |
+
result = upscaler(
|
189 |
+
prompt=prompt,
|
190 |
+
negative_prompt = neg_prompt,
|
191 |
+
image=low_res_latents,
|
192 |
+
num_inference_steps=20,
|
193 |
+
guidance_scale=guidance,
|
194 |
+
generator=generator,
|
195 |
+
)
|
196 |
+
else:
|
197 |
+
result = pipe_i2i(
|
198 |
prompt,
|
199 |
negative_prompt = neg_prompt,
|
200 |
init_image = img,
|
|
|
204 |
#width = width,
|
205 |
#height = height,
|
206 |
generator = generator)
|
207 |
+
|
208 |
|
209 |
return result.images[0]
|
210 |
|
|
|
256 |
neg_prompt = gr.Textbox(label="Negative prompt", placeholder="What to exclude from the image")
|
257 |
disable_auto_prompt_correction = gr.Checkbox(label="Disable auto prompt corretion.")
|
258 |
with gr.Row():
|
259 |
+
image_size=gr.Radio(["Portrait","Landscape","Square","Highreso.","Superreso."])
|
260 |
image_size.show_label=False
|
261 |
image_size.value="Square"
|
262 |
|