Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -27,21 +27,6 @@ device = torch.device("cuda")
|
|
27 |
offload = False
|
28 |
is_schnell = name == "flux-schnell"
|
29 |
|
30 |
-
model, ae, t5, clip, controlnet = None, None, None, None, None
|
31 |
-
|
32 |
-
def load_models():
|
33 |
-
global model, ae, t5, clip, controlnet
|
34 |
-
t5 = load_t5(device, max_length=256 if is_schnell else 512)
|
35 |
-
clip = load_clip(device)
|
36 |
-
model = load_flow_model(name, device=device)
|
37 |
-
ae = load_ae(name, device=device)
|
38 |
-
controlnet = load_controlnet(name, device).to(device).to(torch.bfloat16)
|
39 |
-
|
40 |
-
checkpoint = load_safetensors(model_path)
|
41 |
-
controlnet.load_state_dict(checkpoint, strict=False)
|
42 |
-
|
43 |
-
load_models()
|
44 |
-
|
45 |
def preprocess_image(image, target_width, target_height, crop=True):
|
46 |
if crop:
|
47 |
image = c_crop(image) # Crop the image to square
|
@@ -78,11 +63,16 @@ def generate_image(prompt, control_image, num_steps=50, guidance=4, width=512, h
|
|
78 |
|
79 |
torch_device = torch.device("cuda")
|
80 |
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
|
|
|
|
|
|
|
|
|
|
86 |
|
87 |
width = 16 * width // 16
|
88 |
height = 16 * height // 16
|
@@ -116,8 +106,8 @@ interface = gr.Interface(
|
|
116 |
gr.Image(type="pil", label="Control Image"),
|
117 |
gr.Slider(step=1, minimum=1, maximum=64, value=28, label="Num Steps"),
|
118 |
gr.Slider(minimum=0.1, maximum=10, value=4, label="Guidance"),
|
119 |
-
gr.Slider(minimum=128, maximum=
|
120 |
-
gr.Slider(minimum=128, maximum=
|
121 |
gr.Number(value=42, label="Seed"),
|
122 |
gr.Checkbox(label="Random Seed")
|
123 |
],
|
@@ -127,5 +117,4 @@ interface = gr.Interface(
|
|
127 |
)
|
128 |
|
129 |
if __name__ == "__main__":
|
130 |
-
interface.launch()
|
131 |
-
|
|
|
27 |
offload = False
|
28 |
is_schnell = name == "flux-schnell"
|
29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
def preprocess_image(image, target_width, target_height, crop=True):
|
31 |
if crop:
|
32 |
image = c_crop(image) # Crop the image to square
|
|
|
63 |
|
64 |
torch_device = torch.device("cuda")
|
65 |
|
66 |
+
torch.cuda.empty_cache() # Clear GPU cache
|
67 |
+
|
68 |
+
model = load_flow_model(name, device=torch_device)
|
69 |
+
t5 = load_t5(torch_device, max_length=256 if is_schnell else 512)
|
70 |
+
clip = load_clip(torch_device)
|
71 |
+
ae = load_ae(name, device=torch_device)
|
72 |
+
controlnet = load_controlnet(name, torch_device).to(torch_device).to(torch.bfloat16)
|
73 |
+
|
74 |
+
checkpoint = load_safetensors(model_path)
|
75 |
+
controlnet.load_state_dict(checkpoint, strict=False)
|
76 |
|
77 |
width = 16 * width // 16
|
78 |
height = 16 * height // 16
|
|
|
106 |
gr.Image(type="pil", label="Control Image"),
|
107 |
gr.Slider(step=1, minimum=1, maximum=64, value=28, label="Num Steps"),
|
108 |
gr.Slider(minimum=0.1, maximum=10, value=4, label="Guidance"),
|
109 |
+
gr.Slider(minimum=128, maximum=1024, step=128, value=512, label="Width"),
|
110 |
+
gr.Slider(minimum=128, maximum=1024, step=128, value=512, label="Height"),
|
111 |
gr.Number(value=42, label="Seed"),
|
112 |
gr.Checkbox(label="Random Seed")
|
113 |
],
|
|
|
117 |
)
|
118 |
|
119 |
if __name__ == "__main__":
|
120 |
+
interface.launch()
|
|