Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -8,67 +8,45 @@ import logging
|
|
8 |
import re
|
9 |
import torch
|
10 |
from transformers import GPT2LMHeadModel, GPT2Tokenizer
|
11 |
-
import warnings
|
12 |
|
13 |
-
# Configuraci贸n
|
14 |
logging.basicConfig(level=logging.INFO)
|
15 |
logger = logging.getLogger(__name__)
|
16 |
|
17 |
-
#
|
18 |
-
|
19 |
-
warnings.filterwarnings("ignore", category=DeprecationWarning)
|
20 |
|
21 |
-
|
|
|
22 |
|
23 |
-
#
|
24 |
-
|
25 |
-
|
26 |
-
"es-MX-JorgeNeural", "es-ES-AlvaroNeural", "es-AR-TomasNeural",
|
27 |
-
"en-US-JennyNeural", "fr-FR-DeniseNeural", "de-DE-KatjaNeural"
|
28 |
-
]
|
29 |
-
|
30 |
-
# Cargar modelo GPT-2 con configuraci贸n optimizada
|
31 |
-
try:
|
32 |
-
tokenizer = GPT2Tokenizer.from_pretrained("datificate/gpt2-small-spanish")
|
33 |
-
model = GPT2LMHeadModel.from_pretrained("datificate/gpt2-small-spanish")
|
34 |
-
logger.info("Modelo GPT-2 cargado correctamente")
|
35 |
-
except Exception as e:
|
36 |
-
logger.error(f"Error cargando modelo: {str(e)}")
|
37 |
-
model = None
|
38 |
-
tokenizer = None
|
39 |
|
40 |
def generar_texto(tema):
|
41 |
-
"""Genera texto largo sobre el tema sin estructuras
|
42 |
-
if model is None or tokenizer is None:
|
43 |
-
return f"Contenido sobre {tema}. " * 50
|
44 |
-
|
45 |
try:
|
46 |
-
|
47 |
-
prompt =
|
48 |
|
49 |
-
# Codificar el texto con truncamiento
|
50 |
-
inputs = tokenizer(prompt, return_tensors="pt", max_length=512, truncation=True)
|
51 |
-
|
52 |
-
# Generar texto con par谩metros optimizados
|
53 |
outputs = model.generate(
|
54 |
inputs.input_ids,
|
55 |
max_length=800,
|
56 |
do_sample=True,
|
57 |
temperature=0.7,
|
58 |
-
top_k=
|
59 |
-
num_return_sequences=1,
|
60 |
pad_token_id=tokenizer.eos_token_id
|
61 |
)
|
62 |
|
63 |
texto = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
64 |
return re.sub(r'\s+', ' ', texto).strip()
|
65 |
-
|
66 |
except Exception as e:
|
67 |
-
logger.error(f"Error generando texto: {
|
68 |
-
return f"
|
69 |
|
70 |
-
def
|
71 |
-
"""
|
72 |
try:
|
73 |
headers = {"Authorization": PEXELS_API_KEY}
|
74 |
response = requests.get(
|
@@ -78,64 +56,65 @@ def obtener_videos(tema):
|
|
78 |
)
|
79 |
return response.json().get("videos", [])[:3]
|
80 |
except Exception as e:
|
81 |
-
logger.error(f"Error
|
82 |
return []
|
83 |
|
84 |
-
def crear_video(
|
|
|
85 |
try:
|
86 |
# 1. Generar texto
|
87 |
-
texto = generar_texto(
|
88 |
-
logger.info(f"Texto generado: {len(texto)} caracteres")
|
89 |
|
90 |
-
# 2.
|
91 |
-
|
92 |
subprocess.run([
|
93 |
'edge-tts',
|
94 |
'--voice', voz_seleccionada,
|
95 |
'--text', texto,
|
96 |
-
'--write-media',
|
97 |
], check=True)
|
98 |
|
99 |
-
|
100 |
-
|
|
|
101 |
|
102 |
-
#
|
103 |
-
videos =
|
104 |
clips = []
|
105 |
|
106 |
-
for i, video in enumerate(videos):
|
107 |
try:
|
108 |
-
|
109 |
-
|
110 |
-
temp_file = f"temp_{i}.mp4"
|
111 |
|
112 |
-
# Descargar video
|
113 |
-
|
|
|
114 |
r.raise_for_status()
|
115 |
with open(temp_file, 'wb') as f:
|
116 |
for chunk in r.iter_content(chunk_size=8192):
|
117 |
f.write(chunk)
|
118 |
|
119 |
-
#
|
120 |
clip = VideoFileClip(temp_file)
|
121 |
-
|
122 |
-
clips.append(clip.subclip(0,
|
123 |
|
124 |
except Exception as e:
|
125 |
-
logger.error(f"Error procesando video {i}: {
|
126 |
|
127 |
-
#
|
128 |
if not clips:
|
129 |
-
|
130 |
else:
|
131 |
-
|
132 |
|
133 |
-
|
134 |
|
135 |
-
#
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
fps=24,
|
140 |
codec="libx264",
|
141 |
audio_codec="aac",
|
@@ -143,40 +122,37 @@ def crear_video(prompt, voz_seleccionada):
|
|
143 |
preset='fast'
|
144 |
)
|
145 |
|
146 |
-
return
|
147 |
-
|
148 |
except Exception as e:
|
149 |
-
logger.error(f"Error cr铆tico: {
|
150 |
return None
|
|
|
151 |
finally:
|
152 |
# Limpieza de archivos temporales
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
|
160 |
-
# Interfaz
|
161 |
with gr.Blocks() as app:
|
|
|
|
|
162 |
with gr.Row():
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
video = gr.Video(label="Resultado")
|
170 |
-
|
171 |
btn.click(
|
172 |
fn=crear_video,
|
173 |
inputs=[tema, voz],
|
174 |
-
outputs=
|
175 |
)
|
176 |
|
177 |
if __name__ == "__main__":
|
178 |
-
app.launch(
|
179 |
-
server_name="0.0.0.0",
|
180 |
-
server_port=7860,
|
181 |
-
share=False
|
182 |
-
)
|
|
|
8 |
import re
|
9 |
import torch
|
10 |
from transformers import GPT2LMHeadModel, GPT2Tokenizer
|
|
|
11 |
|
12 |
+
# Configuraci贸n b谩sica
|
13 |
logging.basicConfig(level=logging.INFO)
|
14 |
logger = logging.getLogger(__name__)
|
15 |
|
16 |
+
# Configuraci贸n de entorno (usa tu propia API key de Pexels)
|
17 |
+
PEXELS_API_KEY = os.getenv("PEXELS_API_KEY") or "TU_API_KEY_AQUI"
|
|
|
18 |
|
19 |
+
# Voces disponibles (Edge-TTS)
|
20 |
+
VOICES = ["es-MX-DaliaNeural", "es-ES-ElviraNeural", "en-US-JennyNeural"]
|
21 |
|
22 |
+
# Carga el modelo GPT-2 en espa帽ol (ligero y r谩pido)
|
23 |
+
tokenizer = GPT2Tokenizer.from_pretrained("datificate/gpt2-small-spanish")
|
24 |
+
model = GPT2LMHeadModel.from_pretrained("datificate/gpt2-small-spanish")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
|
26 |
def generar_texto(tema):
|
27 |
+
"""Genera un texto largo y natural sobre el tema (sin estructuras forzadas)."""
|
|
|
|
|
|
|
28 |
try:
|
29 |
+
prompt = f"Habla extensamente sobre {tema} en un tono natural y detallado:"
|
30 |
+
inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
|
31 |
|
|
|
|
|
|
|
|
|
32 |
outputs = model.generate(
|
33 |
inputs.input_ids,
|
34 |
max_length=800,
|
35 |
do_sample=True,
|
36 |
temperature=0.7,
|
37 |
+
top_k=50,
|
|
|
38 |
pad_token_id=tokenizer.eos_token_id
|
39 |
)
|
40 |
|
41 |
texto = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
42 |
return re.sub(r'\s+', ' ', texto).strip()
|
43 |
+
|
44 |
except Exception as e:
|
45 |
+
logger.error(f"Error generando texto: {e}")
|
46 |
+
return f"Contenido generado sobre {tema}."
|
47 |
|
48 |
+
def buscar_videos(tema):
|
49 |
+
"""Busca videos en Pexels y devuelve los 3 m谩s relevantes."""
|
50 |
try:
|
51 |
headers = {"Authorization": PEXELS_API_KEY}
|
52 |
response = requests.get(
|
|
|
56 |
)
|
57 |
return response.json().get("videos", [])[:3]
|
58 |
except Exception as e:
|
59 |
+
logger.error(f"Error buscando videos: {e}")
|
60 |
return []
|
61 |
|
62 |
+
def crear_video(tema, voz_seleccionada):
|
63 |
+
"""Genera el video final con voz y clips de video."""
|
64 |
try:
|
65 |
# 1. Generar texto
|
66 |
+
texto = generar_texto(tema)
|
|
|
67 |
|
68 |
+
# 2. Convertir texto a voz (Edge-TTS)
|
69 |
+
voz_archivo = "narracion.mp3"
|
70 |
subprocess.run([
|
71 |
'edge-tts',
|
72 |
'--voice', voz_seleccionada,
|
73 |
'--text', texto,
|
74 |
+
'--write-media', voz_archivo
|
75 |
], check=True)
|
76 |
|
77 |
+
# 3. Procesar audio
|
78 |
+
audio = AudioFileClip(voz_archivo)
|
79 |
+
duracion_total = audio.duration
|
80 |
|
81 |
+
# 4. Buscar y descargar videos
|
82 |
+
videos = buscar_videos(tema) or buscar_videos("nature")
|
83 |
clips = []
|
84 |
|
85 |
+
for i, video in enumerate(videos[:3]): # M谩ximo 3 videos
|
86 |
try:
|
87 |
+
mejor_calidad = max(video['video_files'], key=lambda x: x.get('width', 0))
|
88 |
+
url_video = mejor_calidad['link']
|
|
|
89 |
|
90 |
+
# Descargar video temporal
|
91 |
+
temp_file = f"temp_video_{i}.mp4"
|
92 |
+
with requests.get(url_video, stream=True) as r:
|
93 |
r.raise_for_status()
|
94 |
with open(temp_file, 'wb') as f:
|
95 |
for chunk in r.iter_content(chunk_size=8192):
|
96 |
f.write(chunk)
|
97 |
|
98 |
+
# Ajustar duraci贸n del clip
|
99 |
clip = VideoFileClip(temp_file)
|
100 |
+
duracion_clip = min(duracion_total / len(videos), clip.duration)
|
101 |
+
clips.append(clip.subclip(0, duracion_clip))
|
102 |
|
103 |
except Exception as e:
|
104 |
+
logger.error(f"Error procesando video {i}: {e}")
|
105 |
|
106 |
+
# 5. Combinar clips (o usar fondo negro si no hay videos)
|
107 |
if not clips:
|
108 |
+
video_final = ColorClip((1280, 720), (0, 0, 0), duration=duracion_total)
|
109 |
else:
|
110 |
+
video_final = concatenate_videoclips(clips, method="compose")
|
111 |
|
112 |
+
video_final = video_final.set_audio(audio)
|
113 |
|
114 |
+
# 6. Exportar video
|
115 |
+
nombre_archivo = f"video_final_{datetime.now().strftime('%Y%m%d_%H%M%S')}.mp4"
|
116 |
+
video_final.write_videofile(
|
117 |
+
nombre_archivo,
|
118 |
fps=24,
|
119 |
codec="libx264",
|
120 |
audio_codec="aac",
|
|
|
122 |
preset='fast'
|
123 |
)
|
124 |
|
125 |
+
return nombre_archivo
|
126 |
+
|
127 |
except Exception as e:
|
128 |
+
logger.error(f"Error cr铆tico: {e}")
|
129 |
return None
|
130 |
+
|
131 |
finally:
|
132 |
# Limpieza de archivos temporales
|
133 |
+
if os.path.exists(voz_archivo):
|
134 |
+
os.remove(voz_archivo)
|
135 |
+
for i in range(3):
|
136 |
+
temp_file = f"temp_video_{i}.mp4"
|
137 |
+
if os.path.exists(temp_file):
|
138 |
+
os.remove(temp_file)
|
139 |
|
140 |
+
# Interfaz de Gradio (sencilla y funcional)
|
141 |
with gr.Blocks() as app:
|
142 |
+
gr.Markdown("# 馃幀 Generador Autom谩tico de Videos")
|
143 |
+
|
144 |
with gr.Row():
|
145 |
+
tema = gr.Textbox(label="Tema del video", placeholder="Ej: 'Historia de la inteligencia artificial'")
|
146 |
+
voz = gr.Dropdown(label="Voz", choices=VOICES, value=VOICES[0])
|
147 |
+
btn = gr.Button("Generar Video", variant="primary")
|
148 |
+
|
149 |
+
salida = gr.Video(label="Resultado")
|
150 |
+
|
|
|
|
|
151 |
btn.click(
|
152 |
fn=crear_video,
|
153 |
inputs=[tema, voz],
|
154 |
+
outputs=salida
|
155 |
)
|
156 |
|
157 |
if __name__ == "__main__":
|
158 |
+
app.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
|
|
|
|