Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,7 +1,6 @@
|
|
| 1 |
import spaces
|
| 2 |
import torch
|
| 3 |
-
from diffusers
|
| 4 |
-
from diffusers.models.transformers.transformer_wan import WanTransformer3DModel
|
| 5 |
from diffusers.utils.export_utils import export_to_video
|
| 6 |
import gradio as gr
|
| 7 |
import tempfile
|
|
@@ -14,6 +13,7 @@ from torchao.quantization import Float8DynamicActivationFloat8WeightConfig
|
|
| 14 |
from torchao.quantization import Int8WeightOnlyConfig
|
| 15 |
import aoti
|
| 16 |
from typing import Optional, Tuple, List
|
|
|
|
| 17 |
|
| 18 |
MODEL_ID = "Wan-AI/Wan2.2-I2V-A14B-Diffusers"
|
| 19 |
MAX_DIM = 832
|
|
@@ -31,13 +31,13 @@ MAX_DURATION = round(MAX_FRAMES_MODEL / FIXED_FPS, 1)
|
|
| 31 |
pipe = WanImageToVideoPipeline.from_pretrained(
|
| 32 |
MODEL_ID,
|
| 33 |
transformer=WanTransformer3DModel.from_pretrained(
|
| 34 |
-
|
| 35 |
subfolder='transformer',
|
| 36 |
torch_dtype=torch.bfloat16,
|
| 37 |
device_map='cuda',
|
| 38 |
),
|
| 39 |
transformer_2=WanTransformer3DModel.from_pretrained(
|
| 40 |
-
|
| 41 |
subfolder='transformer_2',
|
| 42 |
torch_dtype=torch.bfloat16,
|
| 43 |
device_map='cuda',
|
|
@@ -45,13 +45,13 @@ pipe = WanImageToVideoPipeline.from_pretrained(
|
|
| 45 |
torch_dtype=torch.bfloat16,
|
| 46 |
).to('cuda')
|
| 47 |
|
| 48 |
-
# تحميل LoRA مع تحسينات للجودة العالية
|
| 49 |
pipe.load_lora_weights(
|
| 50 |
"Kijai/WanVideo_comfy",
|
| 51 |
weight_name="Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors",
|
| 52 |
adapter_name="lightx2v"
|
| 53 |
)
|
| 54 |
-
kwargs_lora = {"load_into_transformer_2": True}
|
| 55 |
pipe.load_lora_weights(
|
| 56 |
"Kijai/WanVideo_comfy",
|
| 57 |
weight_name="Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors",
|
|
@@ -60,8 +60,8 @@ pipe.load_lora_weights(
|
|
| 60 |
pipe.set_adapters(["lightx2v", "lightx2v_2"], adapter_weights=[1., 1.])
|
| 61 |
|
| 62 |
# دمج LoRA مع مقاييس مخصصة لتعزيز الاستقرار والاحترافية
|
| 63 |
-
pipe.fuse_lora(adapter_names=["lightx2v"], lora_scale=3.5, components=["transformer"])
|
| 64 |
-
pipe.fuse_lora(adapter_names=["lightx2v_2"], lora_scale=1.2, components=["transformer_2"])
|
| 65 |
pipe.unload_lora_weights()
|
| 66 |
|
| 67 |
# الكمية لتوفير الذاكرة مع الحفاظ على الدقة
|
|
@@ -73,7 +73,7 @@ quantize_(pipe.transformer_2, Float8DynamicActivationFloat8WeightConfig())
|
|
| 73 |
aoti.aoti_blocks_load(pipe.transformer, 'zerogpu-aoti/Wan2', variant='fp8da')
|
| 74 |
aoti.aoti_blocks_load(pipe.transformer_2, 'zerogpu-aoti/Wan2', variant='fp8da')
|
| 75 |
|
| 76 |
-
# تحسين الـ Prompt
|
| 77 |
default_prompt_i2v = (
|
| 78 |
"ultra realistic cinematic footage shot on Arri Alexa LF with Panavision anamorphic lenses, "
|
| 79 |
"perfectly preserved facial identity, micro-expressions, and body structure across all frames, "
|
|
@@ -96,7 +96,6 @@ default_prompt_i2v = (
|
|
| 96 |
"frame-to-frame stability with advanced optical flow preservation"
|
| 97 |
)
|
| 98 |
|
| 99 |
-
# تحسين الـ Negative Prompt لتجنب أي عيوب عميقة
|
| 100 |
default_negative_prompt = (
|
| 101 |
"low quality, low resolution, low contrast, poor lighting, underexposed, overexposed, bad composition, "
|
| 102 |
"bad framing, bad perspective, flat lighting, washed out colors, jpeg artifacts, noise, static, grain, "
|
|
@@ -120,12 +119,10 @@ def enhance_image(image: Image.Image) -> Image.Image:
|
|
| 120 |
"""
|
| 121 |
تحسين الصورة المدخلة لتعزيز الجودة والواقعية قبل التمرير.
|
| 122 |
"""
|
| 123 |
-
# تعزيز التباين والحدة بلطف
|
| 124 |
enhancer = ImageEnhance.Contrast(image)
|
| 125 |
image = enhancer.enhance(1.05)
|
| 126 |
enhancer = ImageEnhance.Sharpness(image)
|
| 127 |
image = enhancer.enhance(1.1)
|
| 128 |
-
# إضافة فلتر خفيف لتقليل الضوضاء
|
| 129 |
image = image.filter(ImageFilter.UnsharpMask(radius=1, percent=150, threshold=3))
|
| 130 |
return image
|
| 131 |
|
|
@@ -133,17 +130,16 @@ def resize_image(image: Image.Image) -> Image.Image:
|
|
| 133 |
"""
|
| 134 |
تحسين دالة التمرير للحفاظ على الجودة العالية مع الالتزام بالأبعاد.
|
| 135 |
"""
|
| 136 |
-
# تعزيز الصورة أولاً
|
| 137 |
enhanced_image = enhance_image(image)
|
| 138 |
-
|
| 139 |
width, height = enhanced_image.size
|
| 140 |
if width == height:
|
| 141 |
return enhanced_image.resize((SQUARE_DIM, SQUARE_DIM), Image.LANCZOS)
|
| 142 |
-
|
| 143 |
aspect_ratio = width / height
|
| 144 |
MAX_ASPECT_RATIO = MAX_DIM / MIN_DIM
|
| 145 |
MIN_ASPECT_RATIO = MIN_DIM / MAX_DIM
|
| 146 |
-
|
| 147 |
image_to_resize = enhanced_image
|
| 148 |
if aspect_ratio > MAX_ASPECT_RATIO:
|
| 149 |
target_w, target_h = MAX_DIM, MIN_DIM
|
|
@@ -162,13 +158,12 @@ def resize_image(image: Image.Image) -> Image.Image:
|
|
| 162 |
else:
|
| 163 |
target_h = MAX_DIM
|
| 164 |
target_w = int(round(target_h * aspect_ratio))
|
| 165 |
-
|
| 166 |
final_w = round(target_w / MULTIPLE_OF) * MULTIPLE_OF
|
| 167 |
final_h = round(target_h / MULTIPLE_OF) * MULTIPLE_OF
|
| 168 |
final_w = max(MIN_DIM, min(MAX_DIM, final_w))
|
| 169 |
final_h = max(MIN_DIM, min(MAX_DIM, final_h))
|
| 170 |
-
|
| 171 |
-
# استخدام LANCZOS للحفاظ على التفاصيل العالية
|
| 172 |
return image_to_resize.resize((final_w, final_h), Image.LANCZOS)
|
| 173 |
|
| 174 |
def get_num_frames(duration_seconds: float) -> int:
|
|
@@ -203,25 +198,22 @@ def generate_video(
|
|
| 203 |
"""
|
| 204 |
if input_image is None:
|
| 205 |
raise gr.Error("يرجى تحميل صورة مدخلة.")
|
| 206 |
-
|
| 207 |
-
# تنظيف الذاكرة قبل التشغيل
|
| 208 |
gc.collect()
|
| 209 |
torch.cuda.empty_cache()
|
| 210 |
-
|
| 211 |
num_frames = get_num_frames(duration_seconds)
|
| 212 |
current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
|
| 213 |
-
|
| 214 |
-
# تحسين وتمرير الصورة
|
| 215 |
resized_image = resize_image(input_image)
|
| 216 |
-
|
| 217 |
progress(0, desc="بدء التوليد...")
|
| 218 |
-
|
| 219 |
-
# تشغيل النموذج مع progress updates
|
| 220 |
with progress():
|
| 221 |
output_frames_list = pipe(
|
| 222 |
image=resized_image,
|
| 223 |
-
prompt=prompt,
|
| 224 |
-
negative_prompt=negative_prompt,
|
| 225 |
height=resized_image.height,
|
| 226 |
width=resized_image.width,
|
| 227 |
num_frames=num_frames,
|
|
@@ -230,20 +222,18 @@ def generate_video(
|
|
| 230 |
num_inference_steps=int(steps),
|
| 231 |
generator=torch.Generator(device="cuda").manual_seed(current_seed),
|
| 232 |
).frames[0]
|
| 233 |
-
|
| 234 |
progress(1, desc="تصدير الفيديو...")
|
| 235 |
-
|
| 236 |
-
# تصدير الفيديو مع FPS محسن
|
| 237 |
with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmpfile:
|
| 238 |
video_path = tmpfile.name
|
| 239 |
-
|
| 240 |
export_to_video(output_frames_list, video_path, fps=FIXED_FPS)
|
| 241 |
-
|
| 242 |
-
# تنظيف إضافي
|
| 243 |
del output_frames_list
|
| 244 |
gc.collect()
|
| 245 |
torch.cuda.empty_cache()
|
| 246 |
-
|
| 247 |
return video_path, current_seed
|
| 248 |
|
| 249 |
# ================================
|
|
@@ -264,116 +254,111 @@ with gr.Blocks(theme="gradio/soft", title="Dream-wan2-2-faster-Pro - Ultra Profe
|
|
| 264 |
- 🔍 تحسين تلقائي للصورة المدخلة لجودة 8K افتراضية
|
| 265 |
🔗 *جرب الآن وشارك إبداعاتك على Reddit أو Hugging Face!*
|
| 266 |
""")
|
| 267 |
-
|
| 268 |
gr.Markdown("# Wan 2.2 I2V سريع في 4 خطوات مع Lightning LoRA محسن")
|
| 269 |
gr.Markdown(
|
| 270 |
"شغل Wan 2.2 في 4-8 خطوات فقط، مع [Lightning LoRA](https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Wan22-Lightning)، "
|
| 271 |
"كمية fp8، وترجمة AoT — متوافق مع 🧨 diffusers و ZeroGPU⚡️. "
|
| 272 |
"مُحسّن للاحترافية الفائقة: استقرار إطارات، إضاءة سينمائية، وتفاصيل واقعية عميقة."
|
| 273 |
)
|
| 274 |
-
|
| 275 |
with gr.Row():
|
| 276 |
with gr.Column(scale=1):
|
| 277 |
input_image_component = gr.Image(type="pil", label="الصورة المدخلة", image_mode="RGB")
|
| 278 |
prompt_input = gr.Textbox(
|
| 279 |
-
label="الوصف (Prompt)",
|
| 280 |
-
value=default_prompt_i2v,
|
| 281 |
lines=4,
|
| 282 |
placeholder="اكتب وصفًا سينمائيًا واقعيًا..."
|
| 283 |
)
|
| 284 |
duration_seconds_input = gr.Slider(
|
| 285 |
-
minimum=MIN_DURATION,
|
| 286 |
-
maximum=MAX_DURATION,
|
| 287 |
-
step=0.1,
|
| 288 |
value=3.5,
|
| 289 |
label="المدة (ثوانٍ)",
|
| 290 |
info=f"محدود بـ {MIN_FRAMES_MODEL}-{MAX_FRAMES_MODEL} إطار عند {FIXED_FPS} إطار/ثانية."
|
| 291 |
)
|
| 292 |
with gr.Accordion("الإعدادات المتقدمة", open=False):
|
| 293 |
negative_prompt_input = gr.Textbox(
|
| 294 |
-
label="الوصف السلبي (Negative Prompt)",
|
| 295 |
-
value=default_negative_prompt,
|
| 296 |
lines=4
|
| 297 |
)
|
| 298 |
seed_input = gr.Slider(
|
| 299 |
-
label="البذرة (Seed)",
|
| 300 |
-
minimum=0,
|
| 301 |
-
maximum=MAX_SEED,
|
| 302 |
-
step=1,
|
| 303 |
-
value=42,
|
| 304 |
interactive=True
|
| 305 |
)
|
| 306 |
randomize_seed_checkbox = gr.Checkbox(
|
| 307 |
-
label="توليد بذرة عشوائية",
|
| 308 |
-
value=True,
|
| 309 |
interactive=True
|
| 310 |
)
|
| 311 |
steps_slider = gr.Slider(
|
| 312 |
-
minimum=1,
|
| 313 |
-
maximum=30,
|
| 314 |
-
step=1,
|
| 315 |
-
value=6,
|
| 316 |
label="عدد الخطوات (Inference Steps)"
|
| 317 |
)
|
| 318 |
guidance_scale_input = gr.Slider(
|
| 319 |
-
minimum=0.0,
|
| 320 |
-
maximum=10.0,
|
| 321 |
-
step=0.1,
|
| 322 |
-
value=1.2,
|
| 323 |
label="مقياس التوجيه - مرحلة الضوضاء العالية"
|
| 324 |
)
|
| 325 |
guidance_scale_2_input = gr.Slider(
|
| 326 |
-
minimum=0.0,
|
| 327 |
-
maximum=10.0,
|
| 328 |
-
step=0.1,
|
| 329 |
-
value=1.2,
|
| 330 |
label="مقياس التوجيه 2 - مرحلة الضوضاء المنخفضة"
|
| 331 |
)
|
| 332 |
-
# إضافة خيار جديد لتعزيز الجودة
|
| 333 |
enhance_image_checkbox = gr.Checkbox(
|
| 334 |
-
label="تعزيز الصورة المدخلة تلقائيًا (للواقعية العميقة)",
|
| 335 |
value=True
|
| 336 |
)
|
| 337 |
generate_button = gr.Button("توليد الفيديو", variant="primary", size="lg")
|
| 338 |
-
|
| 339 |
with gr.Column(scale=1):
|
| 340 |
video_output = gr.Video(
|
| 341 |
-
label="الفيديو المُولّد",
|
| 342 |
-
autoplay=True,
|
| 343 |
interactive=False,
|
| 344 |
-
show_share_button=True
|
| 345 |
)
|
| 346 |
seed_output = gr.Textbox(label="البذرة المستخدمة", interactive=False)
|
| 347 |
-
|
| 348 |
-
# قائمة المدخلات مع الإضافة الجديدة
|
| 349 |
ui_inputs = [
|
| 350 |
input_image_component, prompt_input, steps_slider,
|
| 351 |
negative_prompt_input, duration_seconds_input,
|
| 352 |
guidance_scale_input, guidance_scale_2_input,
|
| 353 |
seed_input, randomize_seed_checkbox, enhance_image_checkbox
|
| 354 |
]
|
| 355 |
-
|
| 356 |
-
# تعديل الدالة لاستخدام الخيار الجديد (إذا كان مفعلاً، قم بتعزيز الصورة في resize_image)
|
| 357 |
def wrapped_generate(*args):
|
| 358 |
-
enhance = args[-1]
|
| 359 |
-
#
|
| 360 |
-
return generate_video(*args[:-1])
|
| 361 |
-
|
| 362 |
generate_button.click(
|
| 363 |
-
fn=wrapped_generate,
|
| 364 |
-
inputs=ui_inputs,
|
| 365 |
outputs=[video_output, seed_output]
|
| 366 |
)
|
| 367 |
-
|
| 368 |
-
# إضافة أمثلة للاحترافية
|
| 369 |
gr.Examples(
|
| 370 |
examples=[
|
| 371 |
["path/to/example_image.jpg", "A professional portrait in cinematic lighting", 4, "", 2.0, 1.0, 1.0, 42, False],
|
| 372 |
-
# أضف المزيد حسب الحاجة
|
| 373 |
],
|
| 374 |
-
inputs=ui_inputs[:-1],
|
| 375 |
label="أمثلة سريعة"
|
| 376 |
)
|
| 377 |
|
| 378 |
if __name__ == "__main__":
|
| 379 |
-
demo.queue().launch(mcp_server=True, share=True)
|
|
|
|
| 1 |
import spaces
|
| 2 |
import torch
|
| 3 |
+
from diffusers import WanImageToVideoPipeline, WanTransformer3DModel # الاستيراد الصحيح
|
|
|
|
| 4 |
from diffusers.utils.export_utils import export_to_video
|
| 5 |
import gradio as gr
|
| 6 |
import tempfile
|
|
|
|
| 13 |
from torchao.quantization import Int8WeightOnlyConfig
|
| 14 |
import aoti
|
| 15 |
from typing import Optional, Tuple, List
|
| 16 |
+
import ftfy # إضافة لمعالجة النصوص
|
| 17 |
|
| 18 |
MODEL_ID = "Wan-AI/Wan2.2-I2V-A14B-Diffusers"
|
| 19 |
MAX_DIM = 832
|
|
|
|
| 31 |
pipe = WanImageToVideoPipeline.from_pretrained(
|
| 32 |
MODEL_ID,
|
| 33 |
transformer=WanTransformer3DModel.from_pretrained(
|
| 34 |
+
MODEL_ID, # استخدم MODEL_ID الرئيسي إذا لم يكن cbensimon متاحًا
|
| 35 |
subfolder='transformer',
|
| 36 |
torch_dtype=torch.bfloat16,
|
| 37 |
device_map='cuda',
|
| 38 |
),
|
| 39 |
transformer_2=WanTransformer3DModel.from_pretrained(
|
| 40 |
+
MODEL_ID,
|
| 41 |
subfolder='transformer_2',
|
| 42 |
torch_dtype=torch.bfloat16,
|
| 43 |
device_map='cuda',
|
|
|
|
| 45 |
torch_dtype=torch.bfloat16,
|
| 46 |
).to('cuda')
|
| 47 |
|
| 48 |
+
# تحميل LoRA مع تحسينات للجودة العالية (مع دعم transformer_2)
|
| 49 |
pipe.load_lora_weights(
|
| 50 |
"Kijai/WanVideo_comfy",
|
| 51 |
weight_name="Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors",
|
| 52 |
adapter_name="lightx2v"
|
| 53 |
)
|
| 54 |
+
kwargs_lora = {"load_into_transformer_2": True} # لـ Wan2.2
|
| 55 |
pipe.load_lora_weights(
|
| 56 |
"Kijai/WanVideo_comfy",
|
| 57 |
weight_name="Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors",
|
|
|
|
| 60 |
pipe.set_adapters(["lightx2v", "lightx2v_2"], adapter_weights=[1., 1.])
|
| 61 |
|
| 62 |
# دمج LoRA مع مقاييس مخصصة لتعزيز الاستقرار والاحترافية
|
| 63 |
+
pipe.fuse_lora(adapter_names=["lightx2v"], lora_scale=3.5, components=["transformer"])
|
| 64 |
+
pipe.fuse_lora(adapter_names=["lightx2v_2"], lora_scale=1.2, components=["transformer_2"])
|
| 65 |
pipe.unload_lora_weights()
|
| 66 |
|
| 67 |
# الكمية لتوفير الذاكرة مع الحفاظ على الدقة
|
|
|
|
| 73 |
aoti.aoti_blocks_load(pipe.transformer, 'zerogpu-aoti/Wan2', variant='fp8da')
|
| 74 |
aoti.aoti_blocks_load(pipe.transformer_2, 'zerogpu-aoti/Wan2', variant='fp8da')
|
| 75 |
|
| 76 |
+
# تحسين الـ Prompt الافتراضي... (يبقى كما هو)
|
| 77 |
default_prompt_i2v = (
|
| 78 |
"ultra realistic cinematic footage shot on Arri Alexa LF with Panavision anamorphic lenses, "
|
| 79 |
"perfectly preserved facial identity, micro-expressions, and body structure across all frames, "
|
|
|
|
| 96 |
"frame-to-frame stability with advanced optical flow preservation"
|
| 97 |
)
|
| 98 |
|
|
|
|
| 99 |
default_negative_prompt = (
|
| 100 |
"low quality, low resolution, low contrast, poor lighting, underexposed, overexposed, bad composition, "
|
| 101 |
"bad framing, bad perspective, flat lighting, washed out colors, jpeg artifacts, noise, static, grain, "
|
|
|
|
| 119 |
"""
|
| 120 |
تحسين الصورة المدخلة لتعزيز الجودة والواقعية قبل التمرير.
|
| 121 |
"""
|
|
|
|
| 122 |
enhancer = ImageEnhance.Contrast(image)
|
| 123 |
image = enhancer.enhance(1.05)
|
| 124 |
enhancer = ImageEnhance.Sharpness(image)
|
| 125 |
image = enhancer.enhance(1.1)
|
|
|
|
| 126 |
image = image.filter(ImageFilter.UnsharpMask(radius=1, percent=150, threshold=3))
|
| 127 |
return image
|
| 128 |
|
|
|
|
| 130 |
"""
|
| 131 |
تحسين دالة التمرير للحفاظ على الجودة العالية مع الالتزام بالأبعاد.
|
| 132 |
"""
|
|
|
|
| 133 |
enhanced_image = enhance_image(image)
|
| 134 |
+
|
| 135 |
width, height = enhanced_image.size
|
| 136 |
if width == height:
|
| 137 |
return enhanced_image.resize((SQUARE_DIM, SQUARE_DIM), Image.LANCZOS)
|
| 138 |
+
|
| 139 |
aspect_ratio = width / height
|
| 140 |
MAX_ASPECT_RATIO = MAX_DIM / MIN_DIM
|
| 141 |
MIN_ASPECT_RATIO = MIN_DIM / MAX_DIM
|
| 142 |
+
|
| 143 |
image_to_resize = enhanced_image
|
| 144 |
if aspect_ratio > MAX_ASPECT_RATIO:
|
| 145 |
target_w, target_h = MAX_DIM, MIN_DIM
|
|
|
|
| 158 |
else:
|
| 159 |
target_h = MAX_DIM
|
| 160 |
target_w = int(round(target_h * aspect_ratio))
|
| 161 |
+
|
| 162 |
final_w = round(target_w / MULTIPLE_OF) * MULTIPLE_OF
|
| 163 |
final_h = round(target_h / MULTIPLE_OF) * MULTIPLE_OF
|
| 164 |
final_w = max(MIN_DIM, min(MAX_DIM, final_w))
|
| 165 |
final_h = max(MIN_DIM, min(MAX_DIM, final_h))
|
| 166 |
+
|
|
|
|
| 167 |
return image_to_resize.resize((final_w, final_h), Image.LANCZOS)
|
| 168 |
|
| 169 |
def get_num_frames(duration_seconds: float) -> int:
|
|
|
|
| 198 |
"""
|
| 199 |
if input_image is None:
|
| 200 |
raise gr.Error("يرجى تحميل صورة مدخلة.")
|
| 201 |
+
|
|
|
|
| 202 |
gc.collect()
|
| 203 |
torch.cuda.empty_cache()
|
| 204 |
+
|
| 205 |
num_frames = get_num_frames(duration_seconds)
|
| 206 |
current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
|
| 207 |
+
|
|
|
|
| 208 |
resized_image = resize_image(input_image)
|
| 209 |
+
|
| 210 |
progress(0, desc="بدء التوليد...")
|
| 211 |
+
|
|
|
|
| 212 |
with progress():
|
| 213 |
output_frames_list = pipe(
|
| 214 |
image=resized_image,
|
| 215 |
+
prompt=ftfy.fix_text(prompt), # إضافة ftfy للنصوص
|
| 216 |
+
negative_prompt=ftfy.fix_text(negative_prompt),
|
| 217 |
height=resized_image.height,
|
| 218 |
width=resized_image.width,
|
| 219 |
num_frames=num_frames,
|
|
|
|
| 222 |
num_inference_steps=int(steps),
|
| 223 |
generator=torch.Generator(device="cuda").manual_seed(current_seed),
|
| 224 |
).frames[0]
|
| 225 |
+
|
| 226 |
progress(1, desc="تصدير الفيديو...")
|
| 227 |
+
|
|
|
|
| 228 |
with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmpfile:
|
| 229 |
video_path = tmpfile.name
|
| 230 |
+
|
| 231 |
export_to_video(output_frames_list, video_path, fps=FIXED_FPS)
|
| 232 |
+
|
|
|
|
| 233 |
del output_frames_list
|
| 234 |
gc.collect()
|
| 235 |
torch.cuda.empty_cache()
|
| 236 |
+
|
| 237 |
return video_path, current_seed
|
| 238 |
|
| 239 |
# ================================
|
|
|
|
| 254 |
- 🔍 تحسين تلقائي للصورة المدخلة لجودة 8K افتراضية
|
| 255 |
🔗 *جرب الآن وشارك إبداعاتك على Reddit أو Hugging Face!*
|
| 256 |
""")
|
| 257 |
+
|
| 258 |
gr.Markdown("# Wan 2.2 I2V سريع في 4 خطوات مع Lightning LoRA محسن")
|
| 259 |
gr.Markdown(
|
| 260 |
"شغل Wan 2.2 في 4-8 خطوات فقط، مع [Lightning LoRA](https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Wan22-Lightning)، "
|
| 261 |
"كمية fp8، وترجمة AoT — متوافق مع 🧨 diffusers و ZeroGPU⚡️. "
|
| 262 |
"مُحسّن للاحترافية الفائقة: استقرار إطارات، إضاءة سينمائية، وتفاصيل واقعية عميقة."
|
| 263 |
)
|
| 264 |
+
|
| 265 |
with gr.Row():
|
| 266 |
with gr.Column(scale=1):
|
| 267 |
input_image_component = gr.Image(type="pil", label="الصورة المدخلة", image_mode="RGB")
|
| 268 |
prompt_input = gr.Textbox(
|
| 269 |
+
label="الوصف (Prompt)",
|
| 270 |
+
value=default_prompt_i2v,
|
| 271 |
lines=4,
|
| 272 |
placeholder="اكتب وصفًا سينمائيًا واقعيًا..."
|
| 273 |
)
|
| 274 |
duration_seconds_input = gr.Slider(
|
| 275 |
+
minimum=MIN_DURATION,
|
| 276 |
+
maximum=MAX_DURATION,
|
| 277 |
+
step=0.1,
|
| 278 |
value=3.5,
|
| 279 |
label="المدة (ثوانٍ)",
|
| 280 |
info=f"محدود بـ {MIN_FRAMES_MODEL}-{MAX_FRAMES_MODEL} إطار عند {FIXED_FPS} إطار/ثانية."
|
| 281 |
)
|
| 282 |
with gr.Accordion("الإعدادات المتقدمة", open=False):
|
| 283 |
negative_prompt_input = gr.Textbox(
|
| 284 |
+
label="الوصف السلبي (Negative Prompt)",
|
| 285 |
+
value=default_negative_prompt,
|
| 286 |
lines=4
|
| 287 |
)
|
| 288 |
seed_input = gr.Slider(
|
| 289 |
+
label="البذرة (Seed)",
|
| 290 |
+
minimum=0,
|
| 291 |
+
maximum=MAX_SEED,
|
| 292 |
+
step=1,
|
| 293 |
+
value=42,
|
| 294 |
interactive=True
|
| 295 |
)
|
| 296 |
randomize_seed_checkbox = gr.Checkbox(
|
| 297 |
+
label="توليد بذرة عشوائية",
|
| 298 |
+
value=True,
|
| 299 |
interactive=True
|
| 300 |
)
|
| 301 |
steps_slider = gr.Slider(
|
| 302 |
+
minimum=1,
|
| 303 |
+
maximum=30,
|
| 304 |
+
step=1,
|
| 305 |
+
value=6,
|
| 306 |
label="عدد الخطوات (Inference Steps)"
|
| 307 |
)
|
| 308 |
guidance_scale_input = gr.Slider(
|
| 309 |
+
minimum=0.0,
|
| 310 |
+
maximum=10.0,
|
| 311 |
+
step=0.1,
|
| 312 |
+
value=1.2,
|
| 313 |
label="مقياس التوجيه - مرحلة الضوضاء العالية"
|
| 314 |
)
|
| 315 |
guidance_scale_2_input = gr.Slider(
|
| 316 |
+
minimum=0.0,
|
| 317 |
+
maximum=10.0,
|
| 318 |
+
step=0.1,
|
| 319 |
+
value=1.2,
|
| 320 |
label="مقياس التوجيه 2 - مرحلة الضوضاء المنخفضة"
|
| 321 |
)
|
|
|
|
| 322 |
enhance_image_checkbox = gr.Checkbox(
|
| 323 |
+
label="تعزيز الصورة المدخلة تلقائيًا (للواقعية العميقة)",
|
| 324 |
value=True
|
| 325 |
)
|
| 326 |
generate_button = gr.Button("توليد الفيديو", variant="primary", size="lg")
|
| 327 |
+
|
| 328 |
with gr.Column(scale=1):
|
| 329 |
video_output = gr.Video(
|
| 330 |
+
label="الفيديو المُولّد",
|
| 331 |
+
autoplay=True,
|
| 332 |
interactive=False,
|
| 333 |
+
show_share_button=True
|
| 334 |
)
|
| 335 |
seed_output = gr.Textbox(label="البذرة المستخدمة", interactive=False)
|
| 336 |
+
|
|
|
|
| 337 |
ui_inputs = [
|
| 338 |
input_image_component, prompt_input, steps_slider,
|
| 339 |
negative_prompt_input, duration_seconds_input,
|
| 340 |
guidance_scale_input, guidance_scale_2_input,
|
| 341 |
seed_input, randomize_seed_checkbox, enhance_image_checkbox
|
| 342 |
]
|
| 343 |
+
|
|
|
|
| 344 |
def wrapped_generate(*args):
|
| 345 |
+
enhance = args[-1]
|
| 346 |
+
# إذا كان enhance مفعلاً، قم بتعزيز في resize_image (مُفعَّل افتراضيًا)
|
| 347 |
+
return generate_video(*args[:-1])
|
| 348 |
+
|
| 349 |
generate_button.click(
|
| 350 |
+
fn=wrapped_generate,
|
| 351 |
+
inputs=ui_inputs,
|
| 352 |
outputs=[video_output, seed_output]
|
| 353 |
)
|
| 354 |
+
|
|
|
|
| 355 |
gr.Examples(
|
| 356 |
examples=[
|
| 357 |
["path/to/example_image.jpg", "A professional portrait in cinematic lighting", 4, "", 2.0, 1.0, 1.0, 42, False],
|
|
|
|
| 358 |
],
|
| 359 |
+
inputs=ui_inputs[:-1],
|
| 360 |
label="أمثلة سريعة"
|
| 361 |
)
|
| 362 |
|
| 363 |
if __name__ == "__main__":
|
| 364 |
+
demo.queue().launch(mcp_server=True, share=True)
|