|
|
import gradio as gr |
|
|
import torch |
|
|
from diffusers import SanaVideoPipeline |
|
|
from diffusers.utils import export_to_video |
|
|
import tempfile |
|
|
|
|
|
|
|
|
model_id = "NVLabs/Sana-1.1-0.6B" |
|
|
|
|
|
print("جاري تحميل نموذج Sana الذكي...") |
|
|
|
|
|
pipe = SanaVideoPipeline.from_pretrained( |
|
|
model_id, |
|
|
torch_dtype=torch.float32, |
|
|
use_safetensors=True |
|
|
) |
|
|
|
|
|
|
|
|
pipe.enable_model_cpu_offload() |
|
|
|
|
|
def generate_video(prompt): |
|
|
if not prompt: |
|
|
return None |
|
|
|
|
|
|
|
|
output = pipe( |
|
|
prompt=prompt, |
|
|
height=320, |
|
|
width=320, |
|
|
num_frames=8, |
|
|
num_inference_steps=4, |
|
|
guidance_scale=5.0, |
|
|
).frames[0] |
|
|
|
|
|
|
|
|
temp_file = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False) |
|
|
export_to_video(output, temp_file.name, fps=8) |
|
|
return temp_file.name |
|
|
|
|
|
|
|
|
with gr.Blocks(theme=gr.themes.Soft()) as demo: |
|
|
gr.Markdown("# 🚀 Sana Video: Ultra-Light CPU Gen") |
|
|
gr.Markdown("هذا النموذج يعمل بتقنية Linear Attention لضمان السرعة على المساحات المجانية.") |
|
|
|
|
|
with gr.Row(): |
|
|
with gr.Column(): |
|
|
prompt = gr.Textbox(label="وصف الفيديو", placeholder="A cinematic shot of a golden retriever puppy playing in the snow...") |
|
|
btn = gr.Button("توليد الفيديو (دقيقتين تقريباً)", variant="primary") |
|
|
with gr.Column(): |
|
|
result = gr.Video(label="النتيجة") |
|
|
|
|
|
btn.click(fn=generate_video, inputs=prompt, outputs=result) |
|
|
|
|
|
demo.launch() |
|
|
|
