app.py

import os
import asyncio
import logging
import tempfile
import requests
from datetime import datetime
import edge_tts
import gradio as gr
import torch
from transformers import GPT2Tokenizer, GPT2LMHeadModel
from keybert import KeyBERT
from moviepy.editor import VideoFileClip, concatenate_videoclips, AudioFileClip, CompositeAudioClip
import subprocess
import re
import math
from pydub import AudioSegment
from collections import Counter
import shutil

# Configuración de logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

# Clave API de Pexels
PEXELS_API_KEY = os.environ.get("PEXELS_API_KEY")

# Buscar videos en Pexels usando API REST
def buscar_videos_pexels(query, api_key, per_page=5):
    headers = {"Authorization": api_key}
    try:
        response = requests.get(
            "https://api.pexels.com/videos/search",
            headers=headers,
            params={"query": query, "per_page": per_page, "orientation": "landscape"},
            timeout=15
        )
        response.raise_for_status()
        return response.json().get("videos", [])
    except Exception as e:
        logger.error(f"Error buscando videos en Pexels: {e}")
        return []

# Inicialización de modelos
MODEL_NAME = "datificate/gpt2-small-spanish"  # Modelo en español
try:
    tokenizer = GPT2Tokenizer.from_pretrained(MODEL_NAME)
    model = GPT2LMHeadModel.from_pretrained(MODEL_NAME).eval()
    if tokenizer.pad_token is None:
        tokenizer.pad_token = tokenizer.eos_token
    logger.info("Modelo GPT-2 en español cargado")
except Exception as e:
    logger.error(f"Error al cargar modelo GPT-2: {e}")
    tokenizer = model = None

try:
    kw_model = KeyBERT('distilbert-base-multilingual-cased')  # Modelo multilingüe
    logger.info("KeyBERT cargado")
except Exception as e:
    logger.error(f"Error al cargar KeyBERT: {e}")
    kw_model = None

# Función mejorada para generar guiones
def generate_script(prompt, max_length=150): 
    if not tokenizer or not model:
        return prompt  # Fallback al prompt original
    
    try:
        # Prompt mejorado con instrucciones claras
        enhanced_prompt = f"Escribe un guion corto y coherente sobre: {prompt}"
        
        inputs = tokenizer(enhanced_prompt, return_tensors="pt", truncation=True, max_length=512)
        
        # Parámetros optimizados para español
        outputs = model.generate(
            **inputs,
            max_length=max_length,
            do_sample=True,
            top_p=0.9,
            top_k=40,
            temperature=0.7,
            repetition_penalty=1.5,
            pad_token_id=tokenizer.pad_token_id,
            eos_token_id=tokenizer.eos_token_id
        )
        text = tokenizer.decode(outputs[0], skip_special_tokens=True)
        
        # Limpiar texto generado
        text = re.sub(r'<[^>]+>', '', text)  # Eliminar tokens especiales
        text = text.split(".")[0] + "."  # Tomar la primera oración coherente
        return text
    except Exception as e:
        logger.error(f"Error generando guion: {e}")
        return prompt  # Fallback al prompt original

# Generación de voz
async def text_to_speech(text, output_path, voice="es-ES-ElviraNeural"):
    try:
        communicate = edge_tts.Communicate(text, voice)
        await communicate.save(output_path)
        return True
    except Exception as e:
        logger.error(f"Error en TTS: {e}")
        return False

# Descarga de videos
def download_video_file(url, temp_dir): 
    if not url:
        return None
    
    try:
        response = requests.get(url, stream=True, timeout=30)
        file_name = f"video_{datetime.now().strftime('%H%M%S%f')}.mp4"
        output_path = os.path.join(temp_dir, file_name)
        
        with open(output_path, 'wb') as f:
            for chunk in response.iter_content(chunk_size=8192):
                f.write(chunk)
        return output_path
    except Exception as e:
        logger.error(f"Error descargando video: {e}")
        return None

# Loop para audio
def loop_audio_to_length(audio_clip, target_duration):
    if audio_clip.duration >= target_duration:
        return audio_clip.subclip(0, target_duration)
    
    loops = int(target_duration / audio_clip.duration) + 1
    audios = [audio_clip] * loops
    return concatenate_videoclips(audios).subclip(0, target_duration)

# Extracción de palabras clave robusta
def extract_visual_keywords_from_script(script_text):
    # Limpiar texto
    clean_text = re.sub(r'[^\w\sáéíóúñ]', '', script_text.lower())
    
    # Método 1: KeyBERT si está disponible
    if kw_model:
        try:
            keywords = kw_model.extract_keywords(
                clean_text,
                keyphrase_ngram_range=(1, 1),
                stop_words='spanish',
                top_n=3
            )
            return [kw[0].replace(" ", "+") for kw in keywords]
        except:
            pass  # Fallback al método simple
    
    # Método 2: Frecuencia de palabras (fallback)
    words = clean_text.split()
    stop_words = {"el", "la", "los", "las", "de", "en", "y", "a", "que", "es", "un", "una", "con"}
    keywords = [word for word in words if len(word) > 3 and word not in stop_words]
    
    if not keywords:
        return ["naturaleza"]  # Palabra clave por defecto
    
    # Contar frecuencia y seleccionar las 3 más comunes
    word_counts = Counter(keywords)
    return [word.replace(" ", "+") for word, _ in word_counts.most_common(3)]

# Función principal para crear video
def crear_video(prompt_type, input_text, musica_file=None):
    logger.info(f"Iniciando creación de video: {prompt_type}")
    
    # 1. Generar o usar guion
    if prompt_type == "Generar Guion con IA":
        guion = generate_script(input_text)
    else: 
        guion = input_text
    
    logger.info(f"Guion: {guion[:100]}...")
    
    # Validar guion
    if not guion.strip():
        raise ValueError("El guion está vacío")
    
    # Directorio temporal
    temp_dir = tempfile.mkdtemp()
    temp_files = []
    
    try:
        # 2. Generar audio de voz
        voz_path = os.path.join(temp_dir, "voz.mp3")
        if not asyncio.run(text_to_speech(guion, voz_path)):
            raise ValueError("Error generando voz")
        temp_files.append(voz_path)
        
        audio_tts = AudioFileClip(voz_path)
        audio_duration = audio_tts.duration
        
        # 3. Extraer palabras clave
        keywords = extract_visual_keywords_from_script(guion)
        logger.info(f"Palabras clave: {keywords}")
        
        # 4. Buscar y descargar videos
        videos_data = []
        for keyword in keywords:
            videos_data.extend(buscar_videos_pexels(keyword, PEXELS_API_KEY, per_page=2))
        
        video_paths = []
        for video in videos_data:
            best_quality = max(video['video_files'], key=lambda x: x['width'] * x['height'])
            path = download_video_file(best_quality['link'], temp_dir)
            if path:
                video_paths.append(path)
                temp_files.append(path)
        
        if not video_paths:
            raise ValueError("No se encontraron videos adecuados")
        
        # 5. Procesar videos
        clips = []
        current_duration = 0
        
        for path in video_paths:
            if current_duration >= audio_duration:
                break
                
            try:
                clip = VideoFileClip(path)
                usable_duration = min(clip.duration, 10)
                clips.append(clip.subclip(0, usable_duration))
                current_duration += usable_duration
            except Exception as e:
                logger.warning(f"Error procesando video: {e}")
        
        if not clips:
            raise ValueError("No hay clips válidos")
        
        video_base = concatenate_videoclips(clips, method="compose")
        
        # 6. Manejar música de fondo
        final_audio = audio_tts
        
        if musica_file:
            try:
                # Convertir el archivo de música a formato utilizable
                music_path = os.path.join(temp_dir, "musica.mp3")
                shutil.copyfile(musica_file, music_path)
                temp_files.append(music_path)
                
                musica_audio = AudioFileClip(music_path)
                musica_loop = loop_audio_to_length(musica_audio, audio_duration)
                
                final_audio = CompositeAudioClip([
                    musica_loop.volumex(0.3),
                    audio_tts.volumex(1.0)
                ])
            except Exception as e:
                logger.warning(f"Error procesando música: {e}")
        
        # 7. Crear video final
        video_final = video_base.set_audio(final_audio).subclip(0, audio_duration)
        
        output_path = os.path.join(temp_dir, "final_video.mp4")
        video_final.write_videofile(
            output_path,
            fps=24,
            threads=4,
            codec="libx264",
            audio_codec="aac",
            preset="medium",
            logger=None
        )
        
        return output_path
        
    except Exception as e:
        logger.error(f"Error creando video: {e}")
        raise
    finally:
        # Limpieza
        for path in temp_files:
            try:
                if os.path.isfile(path):
                    os.remove(path)
            except:
                pass
        if os.path.exists(temp_dir):
            shutil.rmtree(temp_dir, ignore_errors=True)

# Función para ejecutar la aplicación
def run_app(prompt_type, prompt_ia, prompt_manual, musica_file):
    input_text = prompt_ia if prompt_type == "Generar Guion con IA" else prompt_manual
    
    if not input_text.strip():
        return None, "Por favor ingresa texto"
    
    try:
        video_path = crear_video(prompt_type, input_text, musica_file)
        return video_path, "✅ Video generado exitosamente"
    except ValueError as ve:
        return None, f"⚠️ Error: {ve}"
    except Exception as e:
        return None, f"❌ Error crítico: {str(e)}"

# Interfaz de Gradio
with gr.Blocks(title="Generador de Videos con IA", theme="soft") as app:
    gr.Markdown("## 🎬 Generador Automático de Videos con IA")
    
    with gr.Tab("Generador de Video"):
        with gr.Row():
            prompt_type = gr.Radio(
                ["Generar Guion con IA", "Usar Mi Guion"], 
                label="Método",
                value="Generar Guion con IA"
            )
        
        with gr.Column(visible=True) as ia_guion_column:
            prompt_ia = gr.Textbox(
                label="Tema para IA", 
                lines=2,
                placeholder="Ej: Un paisaje natural con montañas y ríos..."
            )
        
        with gr.Column(visible=False) as manual_guion_column:
            prompt_manual = gr.Textbox(
                label="Tu Guion Completo", 
                lines=5,
                placeholder="Ej: En este video exploraremos los misterios del océano..."
            )
        
        musica_input = gr.Audio(
            label="Música de fondo (opcional)",
            type="filepath"
        )
        
        boton = gr.Button("✨ Generar Video", variant="primary")
        
        with gr.Column():
            salida_video = gr.Video(label="Video Generado", interactive=False)
            estado_mensaje = gr.Textbox(label="Estado", interactive=False)

    # Manejar visibilidad de columnas
    prompt_type.change(
        lambda x: (gr.update(visible=x == "Generar Guion con IA"), 
                   gr.update(visible=x == "Usar Mi Guion")),
        inputs=prompt_type,
        outputs=[ia_guion_column, manual_guion_column]
    )

    # Lógica de generación
    boton.click(
        lambda: (None, "⏳ Procesando... (puede tardar varios minutos)"),
        outputs=[salida_video, estado_mensaje],
        queue=False
    ).then(
        run_app,
        inputs=[prompt_type, prompt_ia, prompt_manual, musica_input],
        outputs=[salida_video, estado_mensaje]
    )

if __name__ == "__main__":
    app.launch(server_name="0.0.0.0", server_port=7860)