Spaces:
Runtime error
Runtime error
cocktailpeanut
commited on
Commit
•
f1a6530
1
Parent(s):
60d647a
width/height
Browse files
app.py
CHANGED
@@ -51,6 +51,8 @@ def generate_image(
|
|
51 |
prompt,
|
52 |
ckpt,
|
53 |
num_inference_steps,
|
|
|
|
|
54 |
progress=gr.Progress(track_tqdm=True),
|
55 |
mode="sdxl",
|
56 |
):
|
@@ -77,7 +79,7 @@ def generate_image(
|
|
77 |
)
|
78 |
|
79 |
results = pipe(
|
80 |
-
prompt, num_inference_steps=num_inference_steps, guidance_scale=guidance_scale
|
81 |
)
|
82 |
|
83 |
# if SAFETY_CHECKER:
|
@@ -104,7 +106,7 @@ css = """
|
|
104 |
with gr.Blocks(css=css) as demo:
|
105 |
gr.Markdown(
|
106 |
"""
|
107 |
-
# Phased Consistency Model
|
108 |
|
109 |
Phased Consistency Model (PCM) is an image generation technique that addresses the limitations of the Latent Consistency Model (LCM) in high-resolution and text-conditioned image generation.
|
110 |
PCM outperforms LCM across various generation settings and achieves state-of-the-art results in both image and video generation.
|
@@ -118,16 +120,32 @@ PCM outperforms LCM across various generation settings and achieves state-of-the
|
|
118 |
ckpt = gr.Dropdown(
|
119 |
label="Select inference steps",
|
120 |
choices=list(checkpoints.keys()),
|
121 |
-
value="
|
122 |
)
|
123 |
steps = gr.Slider(
|
124 |
label="Number of Inference Steps",
|
125 |
minimum=1,
|
126 |
maximum=20,
|
127 |
step=1,
|
128 |
-
value=
|
129 |
interactive=False,
|
130 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
ckpt.change(
|
132 |
fn=update_steps,
|
133 |
inputs=[ckpt],
|
@@ -169,22 +187,22 @@ PCM outperforms LCM across various generation settings and achieves state-of-the
|
|
169 |
4,
|
170 |
],
|
171 |
],
|
172 |
-
inputs=[prompt, ckpt, steps],
|
173 |
outputs=[img],
|
174 |
fn=generate_image,
|
175 |
-
cache_examples="lazy",
|
176 |
)
|
177 |
|
178 |
gr.on(
|
179 |
fn=generate_image,
|
180 |
triggers=[ckpt.change, prompt.submit, submit_sdxl.click],
|
181 |
-
inputs=[prompt, ckpt, steps],
|
182 |
outputs=[img],
|
183 |
)
|
184 |
gr.on(
|
185 |
fn=lambda *args: generate_image(*args, mode="sd15"),
|
186 |
triggers=[submit_sd15.click],
|
187 |
-
inputs=[prompt, ckpt, steps],
|
188 |
outputs=[img],
|
189 |
)
|
190 |
|
|
|
51 |
prompt,
|
52 |
ckpt,
|
53 |
num_inference_steps,
|
54 |
+
width,
|
55 |
+
height,
|
56 |
progress=gr.Progress(track_tqdm=True),
|
57 |
mode="sdxl",
|
58 |
):
|
|
|
79 |
)
|
80 |
|
81 |
results = pipe(
|
82 |
+
prompt, num_inference_steps=num_inference_steps, guidance_scale=guidance_scale, width=width, height=height
|
83 |
)
|
84 |
|
85 |
# if SAFETY_CHECKER:
|
|
|
106 |
with gr.Blocks(css=css) as demo:
|
107 |
gr.Markdown(
|
108 |
"""
|
109 |
+
# Phased Consistency Model
|
110 |
|
111 |
Phased Consistency Model (PCM) is an image generation technique that addresses the limitations of the Latent Consistency Model (LCM) in high-resolution and text-conditioned image generation.
|
112 |
PCM outperforms LCM across various generation settings and achieves state-of-the-art results in both image and video generation.
|
|
|
120 |
ckpt = gr.Dropdown(
|
121 |
label="Select inference steps",
|
122 |
choices=list(checkpoints.keys()),
|
123 |
+
value="2-Step",
|
124 |
)
|
125 |
steps = gr.Slider(
|
126 |
label="Number of Inference Steps",
|
127 |
minimum=1,
|
128 |
maximum=20,
|
129 |
step=1,
|
130 |
+
value=2,
|
131 |
interactive=False,
|
132 |
)
|
133 |
+
width = gr.Slider(
|
134 |
+
label="Width",
|
135 |
+
minimum=512,
|
136 |
+
maximum=1024,
|
137 |
+
step=256,
|
138 |
+
value=512,
|
139 |
+
interactive=True
|
140 |
+
)
|
141 |
+
height = gr.Slider(
|
142 |
+
label="Height",
|
143 |
+
minimum=512,
|
144 |
+
maximum=1024,
|
145 |
+
step=256,
|
146 |
+
value=512,
|
147 |
+
interactive=True
|
148 |
+
)
|
149 |
ckpt.change(
|
150 |
fn=update_steps,
|
151 |
inputs=[ckpt],
|
|
|
187 |
4,
|
188 |
],
|
189 |
],
|
190 |
+
inputs=[prompt, ckpt, steps, width, height],
|
191 |
outputs=[img],
|
192 |
fn=generate_image,
|
193 |
+
#cache_examples="lazy",
|
194 |
)
|
195 |
|
196 |
gr.on(
|
197 |
fn=generate_image,
|
198 |
triggers=[ckpt.change, prompt.submit, submit_sdxl.click],
|
199 |
+
inputs=[prompt, ckpt, steps, width, height],
|
200 |
outputs=[img],
|
201 |
)
|
202 |
gr.on(
|
203 |
fn=lambda *args: generate_image(*args, mode="sd15"),
|
204 |
triggers=[submit_sd15.click],
|
205 |
+
inputs=[prompt, ckpt, steps, width, height],
|
206 |
outputs=[img],
|
207 |
)
|
208 |
|