Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -3,40 +3,61 @@ import os
|
|
3 |
import subprocess
|
4 |
import tempfile
|
5 |
import glob
|
|
|
6 |
from huggingface_hub import snapshot_download
|
7 |
import gradio as gr
|
|
|
|
|
8 |
|
9 |
-
#
|
10 |
repo_id = "Wan-AI/Wan2.2-TI2V-5B"
|
11 |
print(f"Downloading/loading checkpoints for {repo_id}...")
|
12 |
ckpt_dir = snapshot_download(repo_id, local_dir_use_symlinks=False)
|
13 |
print(f"Using checkpoints from {ckpt_dir}")
|
14 |
|
15 |
-
#
|
16 |
-
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
try:
|
19 |
-
|
20 |
-
|
|
|
|
|
|
|
|
|
21 |
except Exception:
|
22 |
-
|
23 |
-
duration = int(duration_seconds) * int(steps) * 2.25 + 5
|
24 |
-
return duration
|
25 |
-
|
26 |
-
# ---------------- Helper: Find latest .mp4 in cwd ----------------
|
27 |
-
def find_latest_mp4():
|
28 |
-
files = glob.glob("*.mp4")
|
29 |
-
if not files:
|
30 |
-
return None
|
31 |
-
latest_file = max(files, key=os.path.getctime)
|
32 |
-
return latest_file
|
33 |
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
def get_duration0(
|
38 |
-
prompt, size, duration_seconds, steps, progress):
|
39 |
-
"""Calculate dynamic GPU duration based on parameters."""
|
40 |
if duration_seconds >= 3:
|
41 |
return 220
|
42 |
elif steps > 35 and duration_seconds >= 2:
|
@@ -46,17 +67,31 @@ def get_duration0(
|
|
46 |
else:
|
47 |
return 90
|
48 |
|
49 |
-
|
50 |
-
|
|
|
|
|
|
|
|
|
51 |
|
|
|
52 |
|
53 |
-
@spaces.GPU(duration=
|
54 |
def generate_t2v(prompt, size="1280*704", duration_seconds=5, steps=25, progress=gr.Progress(track_tqdm=True)):
|
55 |
if not prompt.strip():
|
56 |
return None, None, "Please enter a prompt."
|
57 |
|
58 |
temp_dir = tempfile.mkdtemp()
|
59 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
cmd = [
|
61 |
"python", "generate.py",
|
62 |
"--task", "ti2v-5B",
|
@@ -75,6 +110,8 @@ def generate_t2v(prompt, size="1280*704", duration_seconds=5, steps=25, progress
|
|
75 |
except subprocess.CalledProcessError as e:
|
76 |
return None, None, f"Error during T2V generation: {e}"
|
77 |
|
|
|
|
|
78 |
video_file = find_latest_mp4()
|
79 |
if video_file is None:
|
80 |
return None, None, "Generation finished but no video file found."
|
@@ -82,7 +119,7 @@ def generate_t2v(prompt, size="1280*704", duration_seconds=5, steps=25, progress
|
|
82 |
dest_path = os.path.join(temp_dir, os.path.basename(video_file))
|
83 |
os.rename(video_file, dest_path)
|
84 |
|
85 |
-
download_link = f"<a href='
|
86 |
return dest_path, download_link, "Text-to-Video generated successfully!"
|
87 |
|
88 |
@spaces.GPU(duration=get_duration)
|
@@ -91,8 +128,17 @@ def generate_i2v(image, prompt, size="1280*704", duration_seconds=5, steps=25, p
|
|
91 |
return None, None, "Please upload an image and enter a prompt."
|
92 |
|
93 |
temp_dir = tempfile.mkdtemp()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
image_path = os.path.join(temp_dir, "input.jpg")
|
95 |
-
image.save(image_path)
|
96 |
|
97 |
cmd = [
|
98 |
"python", "generate.py",
|
@@ -112,6 +158,8 @@ def generate_i2v(image, prompt, size="1280*704", duration_seconds=5, steps=25, p
|
|
112 |
except subprocess.CalledProcessError as e:
|
113 |
return None, None, f"Error during I2V generation: {e}"
|
114 |
|
|
|
|
|
115 |
video_file = find_latest_mp4()
|
116 |
if video_file is None:
|
117 |
return None, None, "Generation finished but no video file found."
|
@@ -119,13 +167,12 @@ def generate_i2v(image, prompt, size="1280*704", duration_seconds=5, steps=25, p
|
|
119 |
dest_path = os.path.join(temp_dir, os.path.basename(video_file))
|
120 |
os.rename(video_file, dest_path)
|
121 |
|
122 |
-
download_link = f"<a href='
|
123 |
return dest_path, download_link, "Image-to-Video generated successfully!"
|
124 |
|
125 |
-
#
|
126 |
-
with gr.Blocks() as demo:
|
127 |
-
|
128 |
|
|
|
129 |
gr.Markdown("## 🎥 Wan2.2-TI2V-5B Video Generator")
|
130 |
gr.Markdown("Choose **Text-to-Video** or **Image-to-Video** mode below.")
|
131 |
|
@@ -134,7 +181,7 @@ with gr.Blocks() as demo:
|
|
134 |
label="Prompt",
|
135 |
value="Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage"
|
136 |
)
|
137 |
-
t2v_size = gr.Textbox(label="Video Size", value="
|
138 |
t2v_duration = gr.Number(label="Video Length (seconds)", value=5)
|
139 |
t2v_steps = gr.Number(label="Inference Steps", value=25)
|
140 |
t2v_btn = gr.Button("Generate from Text")
|
@@ -148,7 +195,7 @@ with gr.Blocks() as demo:
|
|
148 |
)
|
149 |
|
150 |
with gr.Tab("Image-to-Video"):
|
151 |
-
i2v_image = gr.Image(type="
|
152 |
i2v_prompt = gr.Textbox(
|
153 |
label="Prompt",
|
154 |
value=(
|
@@ -160,7 +207,7 @@ with gr.Blocks() as demo:
|
|
160 |
"intricate details and the refreshing atmosphere of the seaside."
|
161 |
)
|
162 |
)
|
163 |
-
i2v_size = gr.Textbox(label="Video Size", value="
|
164 |
i2v_duration = gr.Number(label="Video Length (seconds)", value=5)
|
165 |
i2v_steps = gr.Number(label="Inference Steps", value=25)
|
166 |
i2v_btn = gr.Button("Generate from Image")
|
@@ -173,5 +220,17 @@ with gr.Blocks() as demo:
|
|
173 |
[i2v_video, i2v_download, i2v_status]
|
174 |
)
|
175 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
176 |
if __name__ == "__main__":
|
177 |
demo.launch()
|
|
|
3 |
import subprocess
|
4 |
import tempfile
|
5 |
import glob
|
6 |
+
import gc
|
7 |
from huggingface_hub import snapshot_download
|
8 |
import gradio as gr
|
9 |
+
from PIL import Image
|
10 |
+
import numpy as np
|
11 |
|
12 |
+
# -------- Model Download --------
|
13 |
repo_id = "Wan-AI/Wan2.2-TI2V-5B"
|
14 |
print(f"Downloading/loading checkpoints for {repo_id}...")
|
15 |
ckpt_dir = snapshot_download(repo_id, local_dir_use_symlinks=False)
|
16 |
print(f"Using checkpoints from {ckpt_dir}")
|
17 |
|
18 |
+
# -------- Constants --------
|
19 |
+
FIXED_FPS = 24
|
20 |
+
MIN_FRAMES_MODEL = 8
|
21 |
+
MAX_FRAMES_MODEL = 121
|
22 |
+
MOD_VALUE = 32
|
23 |
+
DEFAULT_H_SLIDER_VALUE = 704
|
24 |
+
DEFAULT_W_SLIDER_VALUE = 1280
|
25 |
+
NEW_FORMULA_MAX_AREA = 1280.0 * 704.0
|
26 |
+
SLIDER_MIN_H, SLIDER_MAX_H = 128, 1280
|
27 |
+
SLIDER_MIN_W, SLIDER_MAX_W = 128, 1280
|
28 |
+
|
29 |
+
# -------- Helpers --------
|
30 |
+
def _calculate_new_dimensions(pil_image):
|
31 |
+
orig_w, orig_h = pil_image.size
|
32 |
+
if orig_w <= 0 or orig_h <= 0:
|
33 |
+
return DEFAULT_H_SLIDER_VALUE, DEFAULT_W_SLIDER_VALUE
|
34 |
+
|
35 |
+
aspect_ratio = orig_h / orig_w
|
36 |
+
calc_h = round(np.sqrt(NEW_FORMULA_MAX_AREA * aspect_ratio))
|
37 |
+
calc_w = round(np.sqrt(NEW_FORMULA_MAX_AREA / aspect_ratio))
|
38 |
+
|
39 |
+
calc_h = max(MOD_VALUE, (calc_h // MOD_VALUE) * MOD_VALUE)
|
40 |
+
calc_w = max(MOD_VALUE, (calc_w // MOD_VALUE) * MOD_VALUE)
|
41 |
+
|
42 |
+
new_h = int(np.clip(calc_h, SLIDER_MIN_H, (SLIDER_MAX_H // MOD_VALUE) * MOD_VALUE))
|
43 |
+
new_w = int(np.clip(calc_w, SLIDER_MIN_W, (SLIDER_MAX_W // MOD_VALUE) * MOD_VALUE))
|
44 |
+
|
45 |
+
return new_h, new_w
|
46 |
+
|
47 |
+
def handle_image_upload_for_dims(uploaded_pil_image, current_h_val, current_w_val):
|
48 |
+
if uploaded_pil_image is None:
|
49 |
+
return gr.update(value=DEFAULT_H_SLIDER_VALUE), gr.update(value=DEFAULT_W_SLIDER_VALUE)
|
50 |
try:
|
51 |
+
if hasattr(uploaded_pil_image, 'shape'):
|
52 |
+
pil_image = Image.fromarray(uploaded_pil_image).convert("RGB")
|
53 |
+
else:
|
54 |
+
pil_image = uploaded_pil_image
|
55 |
+
new_h, new_w = _calculate_new_dimensions(pil_image)
|
56 |
+
return gr.update(value=new_h), gr.update(value=new_w)
|
57 |
except Exception:
|
58 |
+
return gr.update(value=DEFAULT_H_SLIDER_VALUE), gr.update(value=DEFAULT_W_SLIDER_VALUE)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
|
60 |
+
def get_duration(prompt, size, duration_seconds, steps, progress):
|
|
|
|
|
|
|
|
|
|
|
61 |
if duration_seconds >= 3:
|
62 |
return 220
|
63 |
elif steps > 35 and duration_seconds >= 2:
|
|
|
67 |
else:
|
68 |
return 90
|
69 |
|
70 |
+
def find_latest_mp4():
|
71 |
+
files = glob.glob("*.mp4")
|
72 |
+
if not files:
|
73 |
+
return None
|
74 |
+
latest_file = max(files, key=os.path.getctime)
|
75 |
+
return latest_file
|
76 |
|
77 |
+
# -------- Generation Functions --------
|
78 |
|
79 |
+
@spaces.GPU(duration=get_duration)
|
80 |
def generate_t2v(prompt, size="1280*704", duration_seconds=5, steps=25, progress=gr.Progress(track_tqdm=True)):
|
81 |
if not prompt.strip():
|
82 |
return None, None, "Please enter a prompt."
|
83 |
|
84 |
temp_dir = tempfile.mkdtemp()
|
85 |
|
86 |
+
# Ensure size is multiples of MOD_VALUE (h*w)
|
87 |
+
try:
|
88 |
+
h, w = size.lower().replace(" ", "").split("*")
|
89 |
+
h = max(MOD_VALUE, (int(h) // MOD_VALUE) * MOD_VALUE)
|
90 |
+
w = max(MOD_VALUE, (int(w) // MOD_VALUE) * MOD_VALUE)
|
91 |
+
size = f"{h}*{w}"
|
92 |
+
except Exception:
|
93 |
+
size = f"{DEFAULT_H_SLIDER_VALUE}*{DEFAULT_W_SLIDER_VALUE}"
|
94 |
+
|
95 |
cmd = [
|
96 |
"python", "generate.py",
|
97 |
"--task", "ti2v-5B",
|
|
|
110 |
except subprocess.CalledProcessError as e:
|
111 |
return None, None, f"Error during T2V generation: {e}"
|
112 |
|
113 |
+
gc.collect()
|
114 |
+
|
115 |
video_file = find_latest_mp4()
|
116 |
if video_file is None:
|
117 |
return None, None, "Generation finished but no video file found."
|
|
|
119 |
dest_path = os.path.join(temp_dir, os.path.basename(video_file))
|
120 |
os.rename(video_file, dest_path)
|
121 |
|
122 |
+
download_link = f"<a href='{os.path.basename(dest_path)}' download>📥 Download Video</a>"
|
123 |
return dest_path, download_link, "Text-to-Video generated successfully!"
|
124 |
|
125 |
@spaces.GPU(duration=get_duration)
|
|
|
128 |
return None, None, "Please upload an image and enter a prompt."
|
129 |
|
130 |
temp_dir = tempfile.mkdtemp()
|
131 |
+
|
132 |
+
try:
|
133 |
+
h, w = size.lower().replace(" ", "").split("*")
|
134 |
+
h = max(MOD_VALUE, (int(h) // MOD_VALUE) * MOD_VALUE)
|
135 |
+
w = max(MOD_VALUE, (int(w) // MOD_VALUE) * MOD_VALUE)
|
136 |
+
size = f"{h}*{w}"
|
137 |
+
except Exception:
|
138 |
+
size = f"{DEFAULT_H_SLIDER_VALUE}*{DEFAULT_W_SLIDER_VALUE}"
|
139 |
+
|
140 |
image_path = os.path.join(temp_dir, "input.jpg")
|
141 |
+
Image.fromarray(image).save(image_path)
|
142 |
|
143 |
cmd = [
|
144 |
"python", "generate.py",
|
|
|
158 |
except subprocess.CalledProcessError as e:
|
159 |
return None, None, f"Error during I2V generation: {e}"
|
160 |
|
161 |
+
gc.collect()
|
162 |
+
|
163 |
video_file = find_latest_mp4()
|
164 |
if video_file is None:
|
165 |
return None, None, "Generation finished but no video file found."
|
|
|
167 |
dest_path = os.path.join(temp_dir, os.path.basename(video_file))
|
168 |
os.rename(video_file, dest_path)
|
169 |
|
170 |
+
download_link = f"<a href='{os.path.basename(dest_path)}' download>📥 Download Video</a>"
|
171 |
return dest_path, download_link, "Image-to-Video generated successfully!"
|
172 |
|
173 |
+
# -------- Gradio UI --------
|
|
|
|
|
174 |
|
175 |
+
with gr.Blocks() as demo:
|
176 |
gr.Markdown("## 🎥 Wan2.2-TI2V-5B Video Generator")
|
177 |
gr.Markdown("Choose **Text-to-Video** or **Image-to-Video** mode below.")
|
178 |
|
|
|
181 |
label="Prompt",
|
182 |
value="Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage"
|
183 |
)
|
184 |
+
t2v_size = gr.Textbox(label="Video Size (HxW)", value=f"{DEFAULT_H_SLIDER_VALUE}*{DEFAULT_W_SLIDER_VALUE}")
|
185 |
t2v_duration = gr.Number(label="Video Length (seconds)", value=5)
|
186 |
t2v_steps = gr.Number(label="Inference Steps", value=25)
|
187 |
t2v_btn = gr.Button("Generate from Text")
|
|
|
195 |
)
|
196 |
|
197 |
with gr.Tab("Image-to-Video"):
|
198 |
+
i2v_image = gr.Image(type="numpy", label="Upload Image")
|
199 |
i2v_prompt = gr.Textbox(
|
200 |
label="Prompt",
|
201 |
value=(
|
|
|
207 |
"intricate details and the refreshing atmosphere of the seaside."
|
208 |
)
|
209 |
)
|
210 |
+
i2v_size = gr.Textbox(label="Video Size (HxW)", value=f"{DEFAULT_H_SLIDER_VALUE}*{DEFAULT_W_SLIDER_VALUE}")
|
211 |
i2v_duration = gr.Number(label="Video Length (seconds)", value=5)
|
212 |
i2v_steps = gr.Number(label="Inference Steps", value=25)
|
213 |
i2v_btn = gr.Button("Generate from Image")
|
|
|
220 |
[i2v_video, i2v_download, i2v_status]
|
221 |
)
|
222 |
|
223 |
+
# Auto adjust size on image upload for i2v
|
224 |
+
i2v_image.upload(
|
225 |
+
fn=handle_image_upload_for_dims,
|
226 |
+
inputs=[i2v_image, i2v_size, i2v_size],
|
227 |
+
outputs=[i2v_size, i2v_size]
|
228 |
+
)
|
229 |
+
i2v_image.clear(
|
230 |
+
fn=handle_image_upload_for_dims,
|
231 |
+
inputs=[i2v_image, i2v_size, i2v_size],
|
232 |
+
outputs=[i2v_size, i2v_size]
|
233 |
+
)
|
234 |
+
|
235 |
if __name__ == "__main__":
|
236 |
demo.launch()
|