Spaces:
Runtime error
Runtime error
Update stable_cascade.py
Browse files- stable_cascade.py +13 -0
stable_cascade.py
CHANGED
@@ -13,6 +13,7 @@ def generate_images(
|
|
13 |
height=1024,
|
14 |
width=1024,
|
15 |
guidance_scale=4.0,
|
|
|
16 |
num_images_per_prompt=1,
|
17 |
prior_inference_steps=20,
|
18 |
decoder_inference_steps=10
|
@@ -30,10 +31,12 @@ def generate_images(
|
|
30 |
Returns:
|
31 |
- List[PIL.Image]: A list of generated PIL Image objects.
|
32 |
"""
|
|
|
33 |
|
34 |
# Generate image embeddings using the prior model
|
35 |
prior_output = prior(
|
36 |
prompt=prompt,
|
|
|
37 |
height=height,
|
38 |
width=width,
|
39 |
negative_prompt=negative_prompt,
|
@@ -46,6 +49,7 @@ def generate_images(
|
|
46 |
decoder_output = decoder(
|
47 |
image_embeddings=prior_output.image_embeddings.half(),
|
48 |
prompt=prompt,
|
|
|
49 |
negative_prompt=negative_prompt,
|
50 |
guidance_scale=0.0, # Guidance scale typically set to 0 for decoder as guidance is applied in the prior
|
51 |
output_type="pil",
|
@@ -70,6 +74,15 @@ def web_demo():
|
|
70 |
placeholder="Negative Prompt",
|
71 |
show_label=False,
|
72 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
with gr.Row():
|
74 |
with gr.Column():
|
75 |
text2image_num_images_per_prompt = gr.Slider(
|
|
|
13 |
height=1024,
|
14 |
width=1024,
|
15 |
guidance_scale=4.0,
|
16 |
+
seed=42,
|
17 |
num_images_per_prompt=1,
|
18 |
prior_inference_steps=20,
|
19 |
decoder_inference_steps=10
|
|
|
31 |
Returns:
|
32 |
- List[PIL.Image]: A list of generated PIL Image objects.
|
33 |
"""
|
34 |
+
generator = torch.Generator(device="cuda").manual_seed(seed)
|
35 |
|
36 |
# Generate image embeddings using the prior model
|
37 |
prior_output = prior(
|
38 |
prompt=prompt,
|
39 |
+
generator=generator,
|
40 |
height=height,
|
41 |
width=width,
|
42 |
negative_prompt=negative_prompt,
|
|
|
49 |
decoder_output = decoder(
|
50 |
image_embeddings=prior_output.image_embeddings.half(),
|
51 |
prompt=prompt,
|
52 |
+
generator=generator,
|
53 |
negative_prompt=negative_prompt,
|
54 |
guidance_scale=0.0, # Guidance scale typically set to 0 for decoder as guidance is applied in the prior
|
55 |
output_type="pil",
|
|
|
74 |
placeholder="Negative Prompt",
|
75 |
show_label=False,
|
76 |
)
|
77 |
+
|
78 |
+
text2image_num_images_per_prompt = gr.Slider(
|
79 |
+
minimum=1,
|
80 |
+
maximum=1000000,
|
81 |
+
step=10,
|
82 |
+
value=42,
|
83 |
+
label="Seed",
|
84 |
+
)
|
85 |
+
|
86 |
with gr.Row():
|
87 |
with gr.Column():
|
88 |
text2image_num_images_per_prompt = gr.Slider(
|