feat(src): :rocket: First commit

app.py

@@ -1,7 +1,89 @@
 import gradio as gr
+import torch
+from diffusers import HunyuanVideoPipeline, HunyuanVideoTransformer3DModel
+from diffusers.utils import export_to_video
 
-def greet(name):
-    return "Hello " + name + "!!"
+def generate_video(
+    prompt,
+    height=720,
+    width=1280,
+    num_frames=129,
+    num_inference_steps=30,
+    guidance_scale=6.0,
+    flow_shift=7.0
+):
+    """
+    Función para generar video usando HunyuanVideo
+    """
+    try:
+        # Cargar el modelo transformer con bfloat16
+        model_id = "hunyuanvideo-community/HunyuanVideo"
+        transformer = HunyuanVideoTransformer3DModel.from_pretrained(
+            model_id, 
+            subfolder="transformer",
+            torch_dtype=torch.bfloat16
+        )
 
-demo = gr.Interface(fn=greet, inputs="text", outputs="text")
-demo.launch()
+        # Crear el pipeline con float16
+        pipe = HunyuanVideoPipeline.from_pretrained(
+            model_id,
+            transformer=transformer,
+            torch_dtype=torch.float16
+        )
+
+        # Habilitar tiling del VAE para ahorrar memoria
+        pipe.vae.enable_tiling()
+        
+        # Mover el modelo a GPU si está disponible
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        pipe.to(device)
+
+        # Generar el video
+        output = pipe(
+            prompt=prompt,
+            height=height,
+            width=width,
+            num_frames=num_frames,
+            num_inference_steps=num_inference_steps,
+            guidance_scale=guidance_scale,
+            flow_shift=flow_shift
+        ).frames[0]
+
+        # Exportar a video
+        video_path = "output.mp4"
+        export_to_video(output, video_path, fps=15)
+        
+        return video_path
+
+    except Exception as e:
+        return str(e)
+
+# Crear la interfaz de Gradio
+demo = gr.Interface(
+    fn=generate_video,
+    inputs=[
+        gr.Textbox(label="Prompt", placeholder="Describe el video que quieres generar..."),
+        gr.Slider(minimum=320, maximum=1280, value=720, step=16, label="Alto"),
+        gr.Slider(minimum=320, maximum=1280, value=1280, step=16, label="Ancho"),
+        gr.Slider(minimum=61, maximum=129, value=129, step=4, label="Número de frames"),
+        gr.Slider(minimum=1, maximum=50, value=30, label="Pasos de inferencia"),
+        gr.Slider(minimum=1.0, maximum=20.0, value=6.0, label="Guidance Scale"),
+        gr.Slider(minimum=2.0, maximum=12.0, value=7.0, label="Flow Shift")
+    ],
+    outputs=gr.Video(label="Video generado"),
+    title="Generador de Videos con HunyuanVideo",
+    description="""
+    Genera videos a partir de descripciones de texto usando HunyuanVideo.
+    - El prompt debe ser una descripción clara del video que deseas generar
+    - Se recomienda usar resoluciones soportadas (ver tabla en la documentación)
+    - El número de frames debe ser de la forma 4k + 1 (ej: 61, 129)
+    - Flow shift: usar valores bajos (2-5) para resoluciones pequeñas y altos (7-12) para resoluciones grandes
+    """,
+    examples=[
+        ["A cat walks on the grass, realistic style.", 720, 1280, 61, 30, 6.0, 7.0],
+        ["A beautiful sunset over the ocean, cinematic style.", 544, 960, 129, 30, 6.0, 5.0]
+    ]
+)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,6 @@
+torch>=2.0.0
+torchvision>=0.15.0
+diffusers>=0.25.0
+transformers>=4.36.0
+accelerate>=0.25.0
+gradio>=4.0.0