Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -10,7 +10,6 @@ import numpy as np
|
|
| 10 |
from PIL import Image
|
| 11 |
import random
|
| 12 |
import gc
|
| 13 |
-
|
| 14 |
from torchao.quantization import quantize_
|
| 15 |
from torchao.quantization import Float8DynamicActivationFloat8WeightConfig, Int8WeightOnlyConfig
|
| 16 |
import aoti
|
|
@@ -25,13 +24,11 @@ MAX_DIM = 832
|
|
| 25 |
MIN_DIM = 480
|
| 26 |
SQUARE_DIM = 640
|
| 27 |
MULTIPLE_OF = 16
|
| 28 |
-
|
| 29 |
MAX_SEED = np.iinfo(np.int32).max
|
| 30 |
|
| 31 |
FIXED_FPS = 16
|
| 32 |
MIN_FRAMES_MODEL = 8
|
| 33 |
MAX_FRAMES_MODEL = 7720
|
| 34 |
-
|
| 35 |
MIN_DURATION = round(MIN_FRAMES_MODEL / FIXED_FPS, 1)
|
| 36 |
MAX_DURATION = round(MAX_FRAMES_MODEL / FIXED_FPS, 1)
|
| 37 |
|
|
@@ -95,6 +92,7 @@ aoti.aoti_blocks_load(pipe.transformer_2, 'zerogpu-aoti/Wan2', variant='fp8da')
|
|
| 95 |
# DEFAULT PROMPTS
|
| 96 |
# =========================================================
|
| 97 |
default_prompt_i2v = "Make this image come alive with dynamic, cinematic human motion. Create smooth, natural, lifelike animation with fluid transitions, expressive body movement, realistic physics, and elegant camera flow. Deliver a polished, high-quality motion style that feels immersive, artistic, and visually captivating."
|
|
|
|
| 98 |
default_negative_prompt = (
|
| 99 |
"low quality, worst quality, motion artifacts, unstable motion, jitter, frame jitter, wobbling limbs, motion distortion, inconsistent movement, robotic movement, animation-like motion, awkward transitions, incorrect body mechanics, unnatural posing, off-balance poses, broken motion paths, frozen frames, duplicated frames, frame skipping, warped motion, stretching artifacts bad anatomy, incorrect proportions, deformed body, twisted torso, broken joints, dislocated limbs, distorted neck, unnatural spine curvature, malformed hands, extra fingers, missing fingers, fused fingers, distorted legs, extra limbs, collapsed feet, floating feet, foot sliding, foot jitter, backward walking, unnatural gait blurry details, long exposure blur, ghosting, shadow trails, smearing, washed-out colors, overexposure, underexposure, excessive contrast, blown highlights, poorly rendered clothing, fabric glitches, texture warping, clothing merging with body, incorrect cloth physics ugly background, cluttered scene, crowded background, random objects, unwanted text, subtitles, logos, graffiti, grain, noise, static artifacts, compression noise, jpeg artifacts, image-like stillness, painting-like look, cartoon texture, low-resolution textures"
|
| 100 |
)
|
|
@@ -106,13 +104,12 @@ def resize_image(image: Image.Image) -> Image.Image:
|
|
| 106 |
width, height = image.size
|
| 107 |
if width == height:
|
| 108 |
return image.resize((SQUARE_DIM, SQUARE_DIM), Image.LANCZOS)
|
| 109 |
-
|
| 110 |
aspect_ratio = width / height
|
| 111 |
MAX_ASPECT_RATIO = MAX_DIM / MIN_DIM
|
| 112 |
MIN_ASPECT_RATIO = MIN_DIM / MAX_DIM
|
| 113 |
|
| 114 |
image_to_resize = image
|
| 115 |
-
|
| 116 |
if aspect_ratio > MAX_ASPECT_RATIO:
|
| 117 |
crop_width = int(round(height * MAX_ASPECT_RATIO))
|
| 118 |
left = (width - crop_width) // 2
|
|
@@ -121,20 +118,20 @@ def resize_image(image: Image.Image) -> Image.Image:
|
|
| 121 |
crop_height = int(round(width / MIN_ASPECT_RATIO))
|
| 122 |
top = (height - crop_height) // 2
|
| 123 |
image_to_resize = image.crop((0, top, width, top + crop_height))
|
| 124 |
-
|
| 125 |
if width > height:
|
| 126 |
target_w = MAX_DIM
|
| 127 |
target_h = int(round(target_w / aspect_ratio))
|
| 128 |
else:
|
| 129 |
target_h = MAX_DIM
|
| 130 |
target_w = int(round(target_h * aspect_ratio))
|
| 131 |
-
|
| 132 |
final_w = round(target_w / MULTIPLE_OF) * MULTIPLE_OF
|
| 133 |
final_h = round(target_h / MULTIPLE_OF) * MULTIPLE_OF
|
| 134 |
-
|
| 135 |
final_w = max(MIN_DIM, min(MAX_DIM, final_w))
|
| 136 |
final_h = max(MIN_DIM, min(MAX_DIM, final_h))
|
| 137 |
-
|
| 138 |
return image_to_resize.resize((final_w, final_h), Image.LANCZOS)
|
| 139 |
|
| 140 |
# =========================================================
|
|
@@ -152,9 +149,9 @@ def get_duration(
|
|
| 152 |
if input_image is None:
|
| 153 |
return 120 # Return default duration if image is missing to prevent crash
|
| 154 |
# -----------------------------------------------------
|
| 155 |
-
|
| 156 |
BASE_FRAMES_HEIGHT_WIDTH = 81 * 832 * 624
|
| 157 |
BASE_STEP_DURATION = 15
|
|
|
|
| 158 |
width, height = resize_image(input_image).size
|
| 159 |
frames = get_num_frames(duration_seconds)
|
| 160 |
factor = frames * width * height / BASE_FRAMES_HEIGHT_WIDTH
|
|
@@ -179,7 +176,7 @@ def generate_video(
|
|
| 179 |
):
|
| 180 |
if input_image is None:
|
| 181 |
raise gr.Error("Please upload an input image.")
|
| 182 |
-
|
| 183 |
num_frames = get_num_frames(duration_seconds)
|
| 184 |
current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
|
| 185 |
resized_image = resize_image(input_image)
|
|
@@ -199,6 +196,7 @@ def generate_video(
|
|
| 199 |
|
| 200 |
with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmpfile:
|
| 201 |
video_path = tmpfile.name
|
|
|
|
| 202 |
export_to_video(output_frames_list, video_path, fps=FIXED_FPS)
|
| 203 |
return video_path, current_seed
|
| 204 |
|
|
@@ -209,7 +207,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 209 |
|
| 210 |
# --- ADVERTISEMENT BANNER FOR DREAM HUB PRO ---
|
| 211 |
gr.HTML("""
|
| 212 |
-
<div style="background: linear-gradient(90deg, #4f46e5, #9333ea); color: white; padding: 15px; border-radius: 10px; text-align: center; margin-bottom:
|
| 213 |
<div style="display: flex; align-items: center; justify-content: center; gap: 20px; flex-wrap: wrap;">
|
| 214 |
<div style="text-align: left;">
|
| 215 |
<h3 style="margin: 0; font-weight: bold; font-size: 18px;">✨ New: Dream Hub Pro (All-in-One)</h3>
|
|
@@ -223,6 +221,26 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 223 |
</div>
|
| 224 |
</div>
|
| 225 |
""")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 226 |
# ---------------------------------------------
|
| 227 |
|
| 228 |
gr.Markdown("# 🚀 Dream Wan 2.2 Faster Pro (14B) — Ultra Fast I2V with Lightning LoRA")
|
|
@@ -232,12 +250,13 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 232 |
with gr.Column():
|
| 233 |
input_image_component = gr.Image(type="pil", label="Input Image")
|
| 234 |
prompt_input = gr.Textbox(label="Prompt", value=default_prompt_i2v)
|
|
|
|
| 235 |
duration_seconds_input = gr.Slider(
|
| 236 |
minimum=MIN_DURATION, maximum=MAX_DURATION, step=0.1, value=3.5,
|
| 237 |
label="Duration (seconds)",
|
| 238 |
info=f"Model range: {MIN_FRAMES_MODEL}-{MAX_FRAMES_MODEL} frames at {FIXED_FPS}fps."
|
| 239 |
)
|
| 240 |
-
|
| 241 |
with gr.Accordion("Advanced Settings", open=False):
|
| 242 |
negative_prompt_input = gr.Textbox(label="Negative Prompt", value=default_negative_prompt, lines=3)
|
| 243 |
seed_input = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42)
|
|
@@ -245,7 +264,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 245 |
steps_slider = gr.Slider(minimum=1, maximum=30, step=1, value=6, label="Inference Steps")
|
| 246 |
guidance_scale_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1, label="Guidance Scale (high noise)")
|
| 247 |
guidance_scale_2_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1, label="Guidance Scale 2 (low noise)")
|
| 248 |
-
|
| 249 |
generate_button = gr.Button("🎬 Generate Video", variant="primary")
|
| 250 |
|
| 251 |
with gr.Column():
|
|
@@ -257,6 +276,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 257 |
guidance_scale_input, guidance_scale_2_input,
|
| 258 |
seed_input, randomize_seed_checkbox
|
| 259 |
]
|
|
|
|
| 260 |
generate_button.click(fn=generate_video, inputs=ui_inputs, outputs=[video_output, seed_input])
|
| 261 |
|
| 262 |
gr.Examples(
|
|
|
|
| 10 |
from PIL import Image
|
| 11 |
import random
|
| 12 |
import gc
|
|
|
|
| 13 |
from torchao.quantization import quantize_
|
| 14 |
from torchao.quantization import Float8DynamicActivationFloat8WeightConfig, Int8WeightOnlyConfig
|
| 15 |
import aoti
|
|
|
|
| 24 |
MIN_DIM = 480
|
| 25 |
SQUARE_DIM = 640
|
| 26 |
MULTIPLE_OF = 16
|
|
|
|
| 27 |
MAX_SEED = np.iinfo(np.int32).max
|
| 28 |
|
| 29 |
FIXED_FPS = 16
|
| 30 |
MIN_FRAMES_MODEL = 8
|
| 31 |
MAX_FRAMES_MODEL = 7720
|
|
|
|
| 32 |
MIN_DURATION = round(MIN_FRAMES_MODEL / FIXED_FPS, 1)
|
| 33 |
MAX_DURATION = round(MAX_FRAMES_MODEL / FIXED_FPS, 1)
|
| 34 |
|
|
|
|
| 92 |
# DEFAULT PROMPTS
|
| 93 |
# =========================================================
|
| 94 |
default_prompt_i2v = "Make this image come alive with dynamic, cinematic human motion. Create smooth, natural, lifelike animation with fluid transitions, expressive body movement, realistic physics, and elegant camera flow. Deliver a polished, high-quality motion style that feels immersive, artistic, and visually captivating."
|
| 95 |
+
|
| 96 |
default_negative_prompt = (
|
| 97 |
"low quality, worst quality, motion artifacts, unstable motion, jitter, frame jitter, wobbling limbs, motion distortion, inconsistent movement, robotic movement, animation-like motion, awkward transitions, incorrect body mechanics, unnatural posing, off-balance poses, broken motion paths, frozen frames, duplicated frames, frame skipping, warped motion, stretching artifacts bad anatomy, incorrect proportions, deformed body, twisted torso, broken joints, dislocated limbs, distorted neck, unnatural spine curvature, malformed hands, extra fingers, missing fingers, fused fingers, distorted legs, extra limbs, collapsed feet, floating feet, foot sliding, foot jitter, backward walking, unnatural gait blurry details, long exposure blur, ghosting, shadow trails, smearing, washed-out colors, overexposure, underexposure, excessive contrast, blown highlights, poorly rendered clothing, fabric glitches, texture warping, clothing merging with body, incorrect cloth physics ugly background, cluttered scene, crowded background, random objects, unwanted text, subtitles, logos, graffiti, grain, noise, static artifacts, compression noise, jpeg artifacts, image-like stillness, painting-like look, cartoon texture, low-resolution textures"
|
| 98 |
)
|
|
|
|
| 104 |
width, height = image.size
|
| 105 |
if width == height:
|
| 106 |
return image.resize((SQUARE_DIM, SQUARE_DIM), Image.LANCZOS)
|
| 107 |
+
|
| 108 |
aspect_ratio = width / height
|
| 109 |
MAX_ASPECT_RATIO = MAX_DIM / MIN_DIM
|
| 110 |
MIN_ASPECT_RATIO = MIN_DIM / MAX_DIM
|
| 111 |
|
| 112 |
image_to_resize = image
|
|
|
|
| 113 |
if aspect_ratio > MAX_ASPECT_RATIO:
|
| 114 |
crop_width = int(round(height * MAX_ASPECT_RATIO))
|
| 115 |
left = (width - crop_width) // 2
|
|
|
|
| 118 |
crop_height = int(round(width / MIN_ASPECT_RATIO))
|
| 119 |
top = (height - crop_height) // 2
|
| 120 |
image_to_resize = image.crop((0, top, width, top + crop_height))
|
| 121 |
+
|
| 122 |
if width > height:
|
| 123 |
target_w = MAX_DIM
|
| 124 |
target_h = int(round(target_w / aspect_ratio))
|
| 125 |
else:
|
| 126 |
target_h = MAX_DIM
|
| 127 |
target_w = int(round(target_h * aspect_ratio))
|
| 128 |
+
|
| 129 |
final_w = round(target_w / MULTIPLE_OF) * MULTIPLE_OF
|
| 130 |
final_h = round(target_h / MULTIPLE_OF) * MULTIPLE_OF
|
| 131 |
+
|
| 132 |
final_w = max(MIN_DIM, min(MAX_DIM, final_w))
|
| 133 |
final_h = max(MIN_DIM, min(MAX_DIM, final_h))
|
| 134 |
+
|
| 135 |
return image_to_resize.resize((final_w, final_h), Image.LANCZOS)
|
| 136 |
|
| 137 |
# =========================================================
|
|
|
|
| 149 |
if input_image is None:
|
| 150 |
return 120 # Return default duration if image is missing to prevent crash
|
| 151 |
# -----------------------------------------------------
|
|
|
|
| 152 |
BASE_FRAMES_HEIGHT_WIDTH = 81 * 832 * 624
|
| 153 |
BASE_STEP_DURATION = 15
|
| 154 |
+
|
| 155 |
width, height = resize_image(input_image).size
|
| 156 |
frames = get_num_frames(duration_seconds)
|
| 157 |
factor = frames * width * height / BASE_FRAMES_HEIGHT_WIDTH
|
|
|
|
| 176 |
):
|
| 177 |
if input_image is None:
|
| 178 |
raise gr.Error("Please upload an input image.")
|
| 179 |
+
|
| 180 |
num_frames = get_num_frames(duration_seconds)
|
| 181 |
current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
|
| 182 |
resized_image = resize_image(input_image)
|
|
|
|
| 196 |
|
| 197 |
with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmpfile:
|
| 198 |
video_path = tmpfile.name
|
| 199 |
+
|
| 200 |
export_to_video(output_frames_list, video_path, fps=FIXED_FPS)
|
| 201 |
return video_path, current_seed
|
| 202 |
|
|
|
|
| 207 |
|
| 208 |
# --- ADVERTISEMENT BANNER FOR DREAM HUB PRO ---
|
| 209 |
gr.HTML("""
|
| 210 |
+
<div style="background: linear-gradient(90deg, #4f46e5, #9333ea); color: white; padding: 15px; border-radius: 10px; text-align: center; margin-bottom: 10px; box-shadow: 0 4px 15px rgba(0,0,0,0.1);">
|
| 211 |
<div style="display: flex; align-items: center; justify-content: center; gap: 20px; flex-wrap: wrap;">
|
| 212 |
<div style="text-align: left;">
|
| 213 |
<h3 style="margin: 0; font-weight: bold; font-size: 18px;">✨ New: Dream Hub Pro (All-in-One)</h3>
|
|
|
|
| 221 |
</div>
|
| 222 |
</div>
|
| 223 |
""")
|
| 224 |
+
|
| 225 |
+
# --- NEW: PROFESSIONAL YOUTUBE CHANNEL BANNER ---
|
| 226 |
+
gr.HTML("""
|
| 227 |
+
<div style="background: linear-gradient(90deg, #cc0000, #ff0000); color: white; padding: 12px; border-radius: 10px; text-align: center; margin-bottom: 20px; box-shadow: 0 4px 15px rgba(255,0,0,0.2);">
|
| 228 |
+
<div style="display: flex; align-items: center; justify-content: center; gap: 20px; flex-wrap: wrap;">
|
| 229 |
+
<div style="display: flex; align-items: center; gap: 10px;">
|
| 230 |
+
<span style="font-size: 28px;">▶️</span>
|
| 231 |
+
<div style="text-align: left;">
|
| 232 |
+
<h3 style="margin: 0; font-weight: bold; font-size: 18px;">Imagination Engineering</h3>
|
| 233 |
+
<p style="margin: 2px 0 0 0; opacity: 0.9; font-size: 13px;">Tutorials, AI Engineering & Creative Tech</p>
|
| 234 |
+
</div>
|
| 235 |
+
</div>
|
| 236 |
+
<a href="https://www.youtube.com/@ImaginationEngineering" target="_blank" style="text-decoration: none;">
|
| 237 |
+
<button style="background-color: white; color: #cc0000; border: none; padding: 8px 25px; border-radius: 25px; font-weight: bold; cursor: pointer; transition: transform 0.2s; font-size: 15px; box-shadow: 0 2px 5px rgba(0,0,0,0.2);">
|
| 238 |
+
Subscribe & Watch 📺
|
| 239 |
+
</button>
|
| 240 |
+
</a>
|
| 241 |
+
</div>
|
| 242 |
+
</div>
|
| 243 |
+
""")
|
| 244 |
# ---------------------------------------------
|
| 245 |
|
| 246 |
gr.Markdown("# 🚀 Dream Wan 2.2 Faster Pro (14B) — Ultra Fast I2V with Lightning LoRA")
|
|
|
|
| 250 |
with gr.Column():
|
| 251 |
input_image_component = gr.Image(type="pil", label="Input Image")
|
| 252 |
prompt_input = gr.Textbox(label="Prompt", value=default_prompt_i2v)
|
| 253 |
+
|
| 254 |
duration_seconds_input = gr.Slider(
|
| 255 |
minimum=MIN_DURATION, maximum=MAX_DURATION, step=0.1, value=3.5,
|
| 256 |
label="Duration (seconds)",
|
| 257 |
info=f"Model range: {MIN_FRAMES_MODEL}-{MAX_FRAMES_MODEL} frames at {FIXED_FPS}fps."
|
| 258 |
)
|
| 259 |
+
|
| 260 |
with gr.Accordion("Advanced Settings", open=False):
|
| 261 |
negative_prompt_input = gr.Textbox(label="Negative Prompt", value=default_negative_prompt, lines=3)
|
| 262 |
seed_input = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42)
|
|
|
|
| 264 |
steps_slider = gr.Slider(minimum=1, maximum=30, step=1, value=6, label="Inference Steps")
|
| 265 |
guidance_scale_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1, label="Guidance Scale (high noise)")
|
| 266 |
guidance_scale_2_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1, label="Guidance Scale 2 (low noise)")
|
| 267 |
+
|
| 268 |
generate_button = gr.Button("🎬 Generate Video", variant="primary")
|
| 269 |
|
| 270 |
with gr.Column():
|
|
|
|
| 276 |
guidance_scale_input, guidance_scale_2_input,
|
| 277 |
seed_input, randomize_seed_checkbox
|
| 278 |
]
|
| 279 |
+
|
| 280 |
generate_button.click(fn=generate_video, inputs=ui_inputs, outputs=[video_output, seed_input])
|
| 281 |
|
| 282 |
gr.Examples(
|