Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
import gradio as gr
|
2 |
import spaces
|
3 |
-
from clip_slider_pipeline import
|
4 |
from diffusers import FluxPipeline
|
5 |
import torch
|
6 |
import time
|
@@ -22,7 +22,7 @@ def process_controlnet_img(image):
|
|
22 |
pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell",
|
23 |
torch_dtype=torch.bfloat16)
|
24 |
#pipe.enable_model_cpu_offload()
|
25 |
-
|
26 |
|
27 |
base_model = 'black-forest-labs/FLUX.1-schnell'
|
28 |
controlnet_model = 'InstantX/FLUX.1-dev-Controlnet-Canny-alpha'
|
@@ -32,10 +32,9 @@ controlnet_model = 'InstantX/FLUX.1-dev-Controlnet-Canny-alpha'
|
|
32 |
|
33 |
|
34 |
@spaces.GPU(duration=200)
|
35 |
-
def generate(slider_x,
|
36 |
-
x_concept_1, x_concept_2,
|
37 |
avg_diff_x,
|
38 |
-
avg_diff_y,correlation,
|
39 |
img2img_type = None, img = None,
|
40 |
controlnet_scale= None, ip_adapter_scale=None,
|
41 |
|
@@ -47,51 +46,43 @@ def generate(slider_x, slider_y, prompt, seed, iterations, steps, guidance_scale
|
|
47 |
print("x_concept_1", x_concept_1, "x_concept_2", x_concept_2)
|
48 |
|
49 |
if not sorted(slider_x) == sorted([x_concept_1, x_concept_2]):
|
50 |
-
avg_diff =
|
51 |
x_concept_1, x_concept_2 = slider_x[0], slider_x[1]
|
52 |
-
|
53 |
-
|
54 |
-
if not sorted(slider_y) == sorted([y_concept_1, y_concept_2]):
|
55 |
-
avg_diff_2nd = t5_slider.find_latent_direction(slider_y[0], slider_y[1], num_iterations=iterations).to(torch.float16)
|
56 |
-
y_concept_1, y_concept_2 = slider_y[0], slider_y[1]
|
57 |
-
end_time = time.time()
|
58 |
print(f"direction time: {end_time - start_time:.2f} ms")
|
59 |
|
60 |
start_time = time.time()
|
61 |
|
62 |
if img2img_type=="controlnet canny" and img is not None:
|
63 |
control_img = process_controlnet_img(img)
|
64 |
-
image =
|
65 |
elif img2img_type=="ip adapter" and img is not None:
|
66 |
-
image =
|
67 |
else: # text to image
|
68 |
-
image =
|
69 |
|
70 |
end_time = time.time()
|
71 |
print(f"generation time: {end_time - start_time:.2f} ms")
|
72 |
|
73 |
comma_concepts_x = ', '.join(slider_x)
|
74 |
-
comma_concepts_y = ', '.join(slider_y)
|
75 |
|
76 |
avg_diff_x = avg_diff.cpu()
|
77 |
-
avg_diff_y = avg_diff_2nd.cpu()
|
78 |
|
79 |
-
return gr.update(label=comma_concepts_x, interactive=True),
|
80 |
|
81 |
@spaces.GPU
|
82 |
-
def update_scales(x,
|
83 |
-
avg_diff_x,
|
84 |
img2img_type = None, img = None,
|
85 |
controlnet_scale= None, ip_adapter_scale=None,):
|
86 |
avg_diff = avg_diff_x.cuda()
|
87 |
-
avg_diff_2nd = avg_diff_y.cuda()
|
88 |
if img2img_type=="controlnet canny" and img is not None:
|
89 |
control_img = process_controlnet_img(img)
|
90 |
-
image = t5_slider_controlnet.generate(prompt, guidance_scale=guidance_scale, image=control_img, controlnet_conditioning_scale =controlnet_scale, scale=x,
|
91 |
elif img2img_type=="ip adapter" and img is not None:
|
92 |
-
image =
|
93 |
else:
|
94 |
-
image =
|
95 |
return image
|
96 |
|
97 |
|
@@ -103,7 +94,7 @@ def update_x(x,y,prompt,seed, steps,
|
|
103 |
img = None):
|
104 |
avg_diff = avg_diff_x.cuda()
|
105 |
avg_diff_2nd = avg_diff_y.cuda()
|
106 |
-
image =
|
107 |
return image
|
108 |
|
109 |
@spaces.GPU
|
@@ -113,7 +104,7 @@ def update_y(x,y,prompt,seed, steps,
|
|
113 |
img = None):
|
114 |
avg_diff = avg_diff_x.cuda()
|
115 |
avg_diff_2nd = avg_diff_y.cuda()
|
116 |
-
image =
|
117 |
return image
|
118 |
|
119 |
|
@@ -146,29 +137,29 @@ with gr.Blocks(css=css) as demo:
|
|
146 |
|
147 |
x_concept_1 = gr.State("")
|
148 |
x_concept_2 = gr.State("")
|
149 |
-
y_concept_1 = gr.State("")
|
150 |
-
y_concept_2 = gr.State("")
|
151 |
|
152 |
avg_diff_x = gr.State()
|
153 |
-
avg_diff_y = gr.State()
|
154 |
|
155 |
with gr.Tab("text2image"):
|
156 |
with gr.Row():
|
157 |
with gr.Column():
|
158 |
-
slider_x = gr.Dropdown(label="Slider
|
159 |
-
slider_y = gr.Dropdown(label="Slider
|
160 |
prompt = gr.Textbox(label="Prompt")
|
161 |
submit = gr.Button("find directions")
|
162 |
with gr.Column():
|
163 |
with gr.Group(elem_id="group"):
|
164 |
-
x = gr.Slider(minimum=-
|
165 |
-
y = gr.Slider(minimum=-10, value=0, maximum=10, elem_id="y", interactive=False)
|
166 |
output_image = gr.Image(elem_id="image_out")
|
167 |
with gr.Row():
|
168 |
generate_butt = gr.Button("generate")
|
169 |
|
170 |
with gr.Accordion(label="advanced options", open=False):
|
171 |
-
iterations = gr.Slider(label = "num iterations", minimum=0, value=
|
172 |
steps = gr.Slider(label = "num inference steps", minimum=1, value=4, maximum=10)
|
173 |
guidance_scale = gr.Slider(
|
174 |
label="Guidance scale",
|
@@ -177,69 +168,72 @@ with gr.Blocks(css=css) as demo:
|
|
177 |
step=0.1,
|
178 |
value=5,
|
179 |
)
|
180 |
-
correlation = gr.Slider(
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
seed = gr.Slider(minimum=0, maximum=np.iinfo(np.int32).max, label="Seed", interactive=True, randomize=True)
|
188 |
|
189 |
|
190 |
-
with gr.Tab(label="image2image"):
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
|
222 |
-
|
223 |
-
|
224 |
-
|
225 |
-
|
226 |
-
|
227 |
-
|
228 |
-
|
229 |
-
|
230 |
-
|
231 |
-
|
232 |
-
|
233 |
|
|
|
|
|
|
|
234 |
submit.click(fn=generate,
|
235 |
-
inputs=[slider_x,
|
236 |
-
outputs=[x,
|
237 |
-
|
238 |
-
generate_butt.click(fn=update_scales, inputs=[x,
|
239 |
-
generate_butt_a.click(fn=update_scales, inputs=[x_a,y_a, prompt_a, seed_a, steps_a, guidance_scale_a, avg_diff_x, avg_diff_y, img2img_type, image, controlnet_conditioning_scale, ip_adapter_scale], outputs=[output_image_a])
|
240 |
-
submit_a.click(fn=generate,
|
241 |
-
|
242 |
-
|
243 |
|
244 |
|
245 |
if __name__ == "__main__":
|
|
|
1 |
import gradio as gr
|
2 |
import spaces
|
3 |
+
from clip_slider_pipeline import CLIPSliderFlux
|
4 |
from diffusers import FluxPipeline
|
5 |
import torch
|
6 |
import time
|
|
|
22 |
pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell",
|
23 |
torch_dtype=torch.bfloat16)
|
24 |
#pipe.enable_model_cpu_offload()
|
25 |
+
clip_slider = CLIPSliderFlux(pipe, device=torch.device("cuda"))
|
26 |
|
27 |
base_model = 'black-forest-labs/FLUX.1-schnell'
|
28 |
controlnet_model = 'InstantX/FLUX.1-dev-Controlnet-Canny-alpha'
|
|
|
32 |
|
33 |
|
34 |
@spaces.GPU(duration=200)
|
35 |
+
def generate(slider_x, prompt, seed, iterations, steps, guidance_scale,
|
36 |
+
x_concept_1, x_concept_2,
|
37 |
avg_diff_x,
|
|
|
38 |
img2img_type = None, img = None,
|
39 |
controlnet_scale= None, ip_adapter_scale=None,
|
40 |
|
|
|
46 |
print("x_concept_1", x_concept_1, "x_concept_2", x_concept_2)
|
47 |
|
48 |
if not sorted(slider_x) == sorted([x_concept_1, x_concept_2]):
|
49 |
+
avg_diff = clip_slider.find_latent_direction(slider_x[0], slider_x[1], num_iterations=iterations).to(torch.float16)
|
50 |
x_concept_1, x_concept_2 = slider_x[0], slider_x[1]
|
51 |
+
|
|
|
|
|
|
|
|
|
|
|
52 |
print(f"direction time: {end_time - start_time:.2f} ms")
|
53 |
|
54 |
start_time = time.time()
|
55 |
|
56 |
if img2img_type=="controlnet canny" and img is not None:
|
57 |
control_img = process_controlnet_img(img)
|
58 |
+
image = clip_slider.generate(prompt, guidance_scale=guidance_scale, image=control_img, controlnet_conditioning_scale =controlnet_scale, scale=0, scale_2nd=0, seed=seed, num_inference_steps=steps, avg_diff=avg_diff, avg_diff_2nd=avg_diff_2nd)
|
59 |
elif img2img_type=="ip adapter" and img is not None:
|
60 |
+
image = clip_slider.generate(prompt, guidance_scale=guidance_scale, ip_adapter_image=img, scale=0, scale_2nd=0, seed=seed, num_inference_steps=steps, avg_diff=avg_diff, avg_diff_2nd=avg_diff_2nd)
|
61 |
else: # text to image
|
62 |
+
image = clip_slider.generate(prompt, guidance_scale=guidance_scale, scale=0, scale_2nd=0, seed=seed, num_inference_steps=steps, avg_diff=avg_diff)
|
63 |
|
64 |
end_time = time.time()
|
65 |
print(f"generation time: {end_time - start_time:.2f} ms")
|
66 |
|
67 |
comma_concepts_x = ', '.join(slider_x)
|
|
|
68 |
|
69 |
avg_diff_x = avg_diff.cpu()
|
|
|
70 |
|
71 |
+
return gr.update(label=comma_concepts_x, interactive=True), x_concept_1, x_concept_2, avg_diff_x, image
|
72 |
|
73 |
@spaces.GPU
|
74 |
+
def update_scales(x,prompt,seed, steps, guidance_scale,
|
75 |
+
avg_diff_x,
|
76 |
img2img_type = None, img = None,
|
77 |
controlnet_scale= None, ip_adapter_scale=None,):
|
78 |
avg_diff = avg_diff_x.cuda()
|
|
|
79 |
if img2img_type=="controlnet canny" and img is not None:
|
80 |
control_img = process_controlnet_img(img)
|
81 |
+
image = t5_slider_controlnet.generate(prompt, guidance_scale=guidance_scale, image=control_img, controlnet_conditioning_scale =controlnet_scale, scale=x, seed=seed, num_inference_steps=steps, avg_diff=avg_diff)
|
82 |
elif img2img_type=="ip adapter" and img is not None:
|
83 |
+
image = clip_slider.generate(prompt, guidance_scale=guidance_scale, ip_adapter_image=img, scale=x,seed=seed, num_inference_steps=steps, avg_diff=avg_diff)
|
84 |
else:
|
85 |
+
image = clip_slider.generate(prompt, guidance_scale=guidance_scale, scale=x, seed=seed, num_inference_steps=steps, avg_diff=avg_diff)
|
86 |
return image
|
87 |
|
88 |
|
|
|
94 |
img = None):
|
95 |
avg_diff = avg_diff_x.cuda()
|
96 |
avg_diff_2nd = avg_diff_y.cuda()
|
97 |
+
image = clip_slider.generate(prompt, scale=x, scale_2nd=y, seed=seed, num_inference_steps=steps, avg_diff=avg_diff,avg_diff_2nd=avg_diff_2nd)
|
98 |
return image
|
99 |
|
100 |
@spaces.GPU
|
|
|
104 |
img = None):
|
105 |
avg_diff = avg_diff_x.cuda()
|
106 |
avg_diff_2nd = avg_diff_y.cuda()
|
107 |
+
image = clip_slider.generate(prompt, scale=x, scale_2nd=y, seed=seed, num_inference_steps=steps, avg_diff=avg_diff,avg_diff_2nd=avg_diff_2nd)
|
108 |
return image
|
109 |
|
110 |
|
|
|
137 |
|
138 |
x_concept_1 = gr.State("")
|
139 |
x_concept_2 = gr.State("")
|
140 |
+
# y_concept_1 = gr.State("")
|
141 |
+
# y_concept_2 = gr.State("")
|
142 |
|
143 |
avg_diff_x = gr.State()
|
144 |
+
#avg_diff_y = gr.State()
|
145 |
|
146 |
with gr.Tab("text2image"):
|
147 |
with gr.Row():
|
148 |
with gr.Column():
|
149 |
+
slider_x = gr.Dropdown(label="Slider concept range", allow_custom_value=True, multiselect=True, max_choices=2)
|
150 |
+
#slider_y = gr.Dropdown(label="Slider Y concept range", allow_custom_value=True, multiselect=True, max_choices=2)
|
151 |
prompt = gr.Textbox(label="Prompt")
|
152 |
submit = gr.Button("find directions")
|
153 |
with gr.Column():
|
154 |
with gr.Group(elem_id="group"):
|
155 |
+
x = gr.Slider(minimum=-4, value=0, maximum=4, elem_id="x", interactive=False)
|
156 |
+
#y = gr.Slider(minimum=-10, value=0, maximum=10, elem_id="y", interactive=False)
|
157 |
output_image = gr.Image(elem_id="image_out")
|
158 |
with gr.Row():
|
159 |
generate_butt = gr.Button("generate")
|
160 |
|
161 |
with gr.Accordion(label="advanced options", open=False):
|
162 |
+
iterations = gr.Slider(label = "num iterations", minimum=0, value=300, maximum=400)
|
163 |
steps = gr.Slider(label = "num inference steps", minimum=1, value=4, maximum=10)
|
164 |
guidance_scale = gr.Slider(
|
165 |
label="Guidance scale",
|
|
|
168 |
step=0.1,
|
169 |
value=5,
|
170 |
)
|
171 |
+
# correlation = gr.Slider(
|
172 |
+
# label="correlation",
|
173 |
+
# minimum=0.1,
|
174 |
+
# maximum=1.0,
|
175 |
+
# step=0.05,
|
176 |
+
# value=0.6,
|
177 |
+
# )
|
178 |
seed = gr.Slider(minimum=0, maximum=np.iinfo(np.int32).max, label="Seed", interactive=True, randomize=True)
|
179 |
|
180 |
|
181 |
+
# with gr.Tab(label="image2image"):
|
182 |
+
# with gr.Row():
|
183 |
+
# with gr.Column():
|
184 |
+
# image = gr.ImageEditor(type="pil", image_mode="L", crop_size=(512, 512))
|
185 |
+
# slider_x_a = gr.Dropdown(label="Slider X concept range", allow_custom_value=True, multiselect=True, max_choices=2)
|
186 |
+
# slider_y_a = gr.Dropdown(label="Slider X concept range", allow_custom_value=True, multiselect=True, max_choices=2)
|
187 |
+
# img2img_type = gr.Radio(["controlnet canny", "ip adapter"], label="", info="", visible=False, value="controlnet canny")
|
188 |
+
# prompt_a = gr.Textbox(label="Prompt")
|
189 |
+
# submit_a = gr.Button("Submit")
|
190 |
+
# with gr.Column():
|
191 |
+
# with gr.Group(elem_id="group"):
|
192 |
+
# x_a = gr.Slider(minimum=-10, value=0, maximum=10, elem_id="x", interactive=False)
|
193 |
+
# y_a = gr.Slider(minimum=-10, value=0, maximum=10, elem_id="y", interactive=False)
|
194 |
+
# output_image_a = gr.Image(elem_id="image_out")
|
195 |
+
# with gr.Row():
|
196 |
+
# generate_butt_a = gr.Button("generate")
|
197 |
|
198 |
+
# with gr.Accordion(label="advanced options", open=False):
|
199 |
+
# iterations_a = gr.Slider(label = "num iterations", minimum=0, value=200, maximum=300)
|
200 |
+
# steps_a = gr.Slider(label = "num inference steps", minimum=1, value=8, maximum=30)
|
201 |
+
# guidance_scale_a = gr.Slider(
|
202 |
+
# label="Guidance scale",
|
203 |
+
# minimum=0.1,
|
204 |
+
# maximum=10.0,
|
205 |
+
# step=0.1,
|
206 |
+
# value=5,
|
207 |
+
# )
|
208 |
+
# controlnet_conditioning_scale = gr.Slider(
|
209 |
+
# label="controlnet conditioning scale",
|
210 |
+
# minimum=0.5,
|
211 |
+
# maximum=5.0,
|
212 |
+
# step=0.1,
|
213 |
+
# value=0.7,
|
214 |
+
# )
|
215 |
+
# ip_adapter_scale = gr.Slider(
|
216 |
+
# label="ip adapter scale",
|
217 |
+
# minimum=0.5,
|
218 |
+
# maximum=5.0,
|
219 |
+
# step=0.1,
|
220 |
+
# value=0.8,
|
221 |
+
# visible=False
|
222 |
+
# )
|
223 |
+
# seed_a = gr.Slider(minimum=0, maximum=np.iinfo(np.int32).max, label="Seed", interactive=True, randomize=True)
|
224 |
|
225 |
+
# submit.click(fn=generate,
|
226 |
+
# inputs=[slider_x, slider_y, prompt, seed, iterations, steps, guidance_scale, x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x, avg_diff_y],
|
227 |
+
# outputs=[x, y, x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x, avg_diff_y, output_image])
|
228 |
submit.click(fn=generate,
|
229 |
+
inputs=[slider_x, prompt, seed, iterations, steps, guidance_scale, x_concept_1, x_concept_2, avg_diff_x],
|
230 |
+
outputs=[x, x_concept_1, x_concept_2, avg_diff_x, output_image])
|
231 |
+
|
232 |
+
generate_butt.click(fn=update_scales, inputs=[x, prompt, seed, steps, guidance_scale, avg_diff_x], outputs=[output_image])
|
233 |
+
# generate_butt_a.click(fn=update_scales, inputs=[x_a,y_a, prompt_a, seed_a, steps_a, guidance_scale_a, avg_diff_x, avg_diff_y, img2img_type, image, controlnet_conditioning_scale, ip_adapter_scale], outputs=[output_image_a])
|
234 |
+
# submit_a.click(fn=generate,
|
235 |
+
# inputs=[slider_x_a, slider_y_a, prompt_a, seed_a, iterations_a, steps_a, guidance_scale_a, x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x, avg_diff_y, img2img_type, image, controlnet_conditioning_scale, ip_adapter_scale],
|
236 |
+
# outputs=[x_a, y_a, x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x, avg_diff_y, output_image_a])
|
237 |
|
238 |
|
239 |
if __name__ == "__main__":
|