Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -8,15 +8,15 @@ import torch
|
|
8 |
from typing import Optional
|
9 |
|
10 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
11 |
-
|
12 |
|
13 |
if torch.cuda.is_available():
|
14 |
torch_dtype = torch.float16
|
15 |
else:
|
16 |
torch_dtype = torch.float32
|
17 |
|
18 |
-
|
19 |
-
|
20 |
|
21 |
MAX_SEED = np.iinfo(np.int32).max
|
22 |
MAX_IMAGE_SIZE = 1024
|
@@ -36,19 +36,24 @@ def infer(
|
|
36 |
):
|
37 |
generator = torch.Generator().manual_seed(seed)
|
38 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
if model_id != model_repo_id_default:
|
40 |
pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch_dtype)
|
41 |
pipe = pipe.to(device)
|
|
|
|
|
|
|
42 |
|
43 |
-
|
44 |
-
prompt=prompt,
|
45 |
-
negative_prompt=negative_prompt,
|
46 |
-
guidance_scale=guidance_scale,
|
47 |
-
num_inference_steps=num_inference_steps,
|
48 |
-
width=width,
|
49 |
-
height=height,
|
50 |
-
generator=generator,
|
51 |
-
).images[0]
|
52 |
|
53 |
return image, pipe.name_or_path
|
54 |
|
|
|
8 |
from typing import Optional
|
9 |
|
10 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
11 |
+
model_id_default = "CompVis/stable-diffusion-v1-4"
|
12 |
|
13 |
if torch.cuda.is_available():
|
14 |
torch_dtype = torch.float16
|
15 |
else:
|
16 |
torch_dtype = torch.float32
|
17 |
|
18 |
+
pipe_default = DiffusionPipeline.from_pretrained(model_id_default, torch_dtype=torch_dtype)
|
19 |
+
pipe_default = pipe_default.to(device)
|
20 |
|
21 |
MAX_SEED = np.iinfo(np.int32).max
|
22 |
MAX_IMAGE_SIZE = 1024
|
|
|
36 |
):
|
37 |
generator = torch.Generator().manual_seed(seed)
|
38 |
|
39 |
+
params = {
|
40 |
+
'prompt': prompt,
|
41 |
+
'negative_prompt': negative_prompt,
|
42 |
+
'guidance_scale': guidance_scale,
|
43 |
+
'num_inference_steps': num_inference_steps,
|
44 |
+
'width': width,
|
45 |
+
'height': height,
|
46 |
+
'generator': generator,
|
47 |
+
}
|
48 |
+
|
49 |
if model_id != model_repo_id_default:
|
50 |
pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch_dtype)
|
51 |
pipe = pipe.to(device)
|
52 |
+
image = pipe(**params).images[0]
|
53 |
+
else:
|
54 |
+
image = pipe_default(**params).images[0]
|
55 |
|
56 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
return image, pipe.name_or_path
|
59 |
|