Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -8,12 +8,17 @@ import gradio
|
|
8 |
import torch
|
9 |
from diffusers import StableDiffusionPipeline
|
10 |
from torch import autocast
|
|
|
|
|
|
|
|
|
11 |
|
12 |
|
13 |
openai.api_key = os.getenv('openaikey')
|
14 |
|
15 |
def predict(input, manual_query_repacement, history=[]):
|
16 |
|
|
|
17 |
if manual_query_repacement != "":
|
18 |
input = manual_query_repacement
|
19 |
|
@@ -30,15 +35,64 @@ def predict(input, manual_query_repacement, history=[]):
|
|
30 |
responseText = response["choices"][0]["text"]
|
31 |
history.append((input, responseText))
|
32 |
|
33 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
|
36 |
|
37 |
inputText = gradio.Textbox(value="tmp")
|
38 |
manual_query = gradio.Textbox(placeholder="Input any query here, to replace the image generation query builder entirely.")
|
39 |
|
|
|
|
|
|
|
40 |
gradio.Interface(fn=predict,
|
41 |
inputs=[inputText,manual_query,'state'],
|
42 |
|
43 |
-
outputs=["chatbot",'state']).launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
|
|
|
8 |
import torch
|
9 |
from diffusers import StableDiffusionPipeline
|
10 |
from torch import autocast
|
11 |
+
#from PIL import Image
|
12 |
+
#from torchvision import transforms
|
13 |
+
|
14 |
+
#from diffusers import StableDiffusionImageVariationPipeline
|
15 |
|
16 |
|
17 |
openai.api_key = os.getenv('openaikey')
|
18 |
|
19 |
def predict(input, manual_query_repacement, history=[]):
|
20 |
|
21 |
+
# gpt3
|
22 |
if manual_query_repacement != "":
|
23 |
input = manual_query_repacement
|
24 |
|
|
|
35 |
responseText = response["choices"][0]["text"]
|
36 |
history.append((input, responseText))
|
37 |
|
38 |
+
|
39 |
+
#img generation
|
40 |
+
prompt = "Yoda"
|
41 |
+
scale = 10
|
42 |
+
n_samples = 4
|
43 |
+
|
44 |
+
# Sometimes the nsfw checker is confused by the Naruto images, you can disable
|
45 |
+
# it at your own risk here
|
46 |
+
#disable_safety = False
|
47 |
+
|
48 |
+
#if disable_safety:
|
49 |
+
# def null_safety(images, **kwargs):
|
50 |
+
# return images, False
|
51 |
+
# pipe.safety_checker = null_safety
|
52 |
+
|
53 |
+
with autocast("cuda"):
|
54 |
+
images = pipe(n_samples*[prompt], guidance_scale=scale).images
|
55 |
+
|
56 |
+
for idx, im in enumerate(images):
|
57 |
+
im.save(f"{idx:06}.png")
|
58 |
+
|
59 |
+
images_list = pipe(
|
60 |
+
inp.tile(n_samples, 1, 1, 1),
|
61 |
+
guidance_scale=scale,
|
62 |
+
num_inference_steps=steps,
|
63 |
+
generator=generator,
|
64 |
+
)
|
65 |
+
|
66 |
+
images = []
|
67 |
+
for i, image in enumerate(images_list["images"]):
|
68 |
+
if(images_list["nsfw_content_detected"][i]):
|
69 |
+
safe_image = Image.open(r"unsafe.png")
|
70 |
+
images.append(safe_image)
|
71 |
+
else:
|
72 |
+
images.append(image)
|
73 |
+
|
74 |
+
|
75 |
+
|
76 |
+
return history, history, images
|
77 |
|
78 |
|
79 |
|
80 |
inputText = gradio.Textbox(value="tmp")
|
81 |
manual_query = gradio.Textbox(placeholder="Input any query here, to replace the image generation query builder entirely.")
|
82 |
|
83 |
+
output_img = gr.Gallery(label="Generated image")
|
84 |
+
output_img.style(grid=2)
|
85 |
+
|
86 |
gradio.Interface(fn=predict,
|
87 |
inputs=[inputText,manual_query,'state'],
|
88 |
|
89 |
+
outputs=["chatbot",'state', output_img]).launch()
|
90 |
+
|
91 |
+
|
92 |
+
|
93 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
94 |
+
pipe = StableDiffusionPipeline.from_pretrained("stale2000/sd-dnditem", torch_dtype=torch.float16)
|
95 |
+
pipe = pipe.to(device)
|
96 |
+
|
97 |
+
|
98 |
|