Upload app.py with huggingface_hub
Browse files
app.py
CHANGED
|
@@ -16,32 +16,8 @@ import subprocess
|
|
| 16 |
import urllib.request
|
| 17 |
from pathlib import Path
|
| 18 |
|
| 19 |
-
# Force Gradio temp dir configuration
|
| 20 |
-
os.environ["GRADIO_TEMP_DIR"] = "/tmp/gradio"
|
| 21 |
-
Path("/tmp/gradio").mkdir(parents=True, exist_ok=True)
|
| 22 |
-
|
| 23 |
import gradio as gr
|
| 24 |
|
| 25 |
-
# ββ Monkeypatch Gradio Client bug in Python 3.13 / Pydantic v2 βββββββββββββββ
|
| 26 |
-
try:
|
| 27 |
-
import gradio_client.utils
|
| 28 |
-
orig_get_type = gradio_client.utils.get_type
|
| 29 |
-
def patched_get_type(schema):
|
| 30 |
-
if isinstance(schema, bool):
|
| 31 |
-
return "any"
|
| 32 |
-
return orig_get_type(schema)
|
| 33 |
-
gradio_client.utils.get_type = patched_get_type
|
| 34 |
-
|
| 35 |
-
orig_json_schema_to_python_type = gradio_client.utils._json_schema_to_python_type
|
| 36 |
-
def patched_json_schema_to_python_type(schema, defs=None):
|
| 37 |
-
if isinstance(schema, bool):
|
| 38 |
-
return "any"
|
| 39 |
-
return orig_json_schema_to_python_type(schema, defs)
|
| 40 |
-
gradio_client.utils._json_schema_to_python_type = patched_json_schema_to_python_type
|
| 41 |
-
print("[patch] Gradio Client JSON schema parser patched successfully.")
|
| 42 |
-
except Exception as patch_err:
|
| 43 |
-
print(f"[patch] Failed to apply Gradio Client patch: {patch_err}")
|
| 44 |
-
|
| 45 |
# ββ Paths βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 46 |
MODELS_DIR = Path("/tmp/voicekit_models")
|
| 47 |
WAV2LIP_DIR = MODELS_DIR / "Wav2Lip"
|
|
@@ -56,7 +32,7 @@ CKPT_URLS = [
|
|
| 56 |
"https://huggingface.co/numz/wav2lip_studio/resolve/main/Wav2lip/wav2lip_gan.pth",
|
| 57 |
"https://huggingface.co/camenduru/wav2lip/resolve/main/wav2lip_gan.pth",
|
| 58 |
]
|
| 59 |
-
FACE_DET_URL = "https://huggingface.co/camenduru/
|
| 60 |
|
| 61 |
|
| 62 |
# ββ One-time Model Bootstrap ββββββββββββββββββββββββββββββββββββββββββββββββββ
|
|
@@ -107,64 +83,60 @@ def bootstrap():
|
|
| 107 |
# Face detector
|
| 108 |
_download(FACE_DET_URL, FACE_DET, "S3FD face detector (~85MB)")
|
| 109 |
|
| 110 |
-
# Patch audio.py for newer librosa versions
|
| 111 |
-
audio_py = WAV2LIP_DIR / "audio.py"
|
| 112 |
-
if audio_py.exists():
|
| 113 |
-
content = audio_py.read_text(encoding="utf-8", errors="ignore")
|
| 114 |
-
if "hp.sample_rate, hp.n_fft" in content:
|
| 115 |
-
content = content.replace(
|
| 116 |
-
"hp.sample_rate, hp.n_fft",
|
| 117 |
-
"sr=hp.sample_rate, n_fft=hp.n_fft"
|
| 118 |
-
)
|
| 119 |
-
audio_py.write_text(content, encoding="utf-8")
|
| 120 |
-
print("[patch] Patched audio.py for librosa filters.mel signature")
|
| 121 |
-
|
| 122 |
print("[ready] All models ready")
|
| 123 |
|
| 124 |
|
| 125 |
-
#
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 129 |
|
| 130 |
|
| 131 |
# ββ Voice Generation ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 132 |
|
| 133 |
def generate_voice(script: str, voice_sample_path: str) -> str:
|
| 134 |
-
"""Generate speech using
|
| 135 |
-
import
|
| 136 |
-
|
| 137 |
|
| 138 |
out_path = str(OUTPUT_DIR / f"voice_{int(time.time())}.wav")
|
| 139 |
-
|
| 140 |
-
|
| 141 |
try:
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 152 |
return out_path
|
| 153 |
except Exception as e:
|
| 154 |
-
print(f"[tts]
|
| 155 |
-
|
| 156 |
|
| 157 |
|
| 158 |
-
def generate_voice_edgetts(script: str
|
| 159 |
"""Fallback: edge-tts generic voice."""
|
| 160 |
import asyncio
|
| 161 |
import edge_tts
|
| 162 |
out_path = str(OUTPUT_DIR / f"voice_generic_{int(time.time())}.mp3")
|
| 163 |
|
| 164 |
-
edge_voice = "en-GB-SoniaNeural" if accent == "uk" else "en-US-AriaNeural"
|
| 165 |
-
|
| 166 |
async def _gen():
|
| 167 |
-
c = edge_tts.Communicate(script,
|
| 168 |
await c.save(out_path)
|
| 169 |
|
| 170 |
asyncio.run(_gen())
|
|
@@ -175,7 +147,7 @@ def generate_voice_edgetts(script: str, accent: str = "us") -> str:
|
|
| 175 |
|
| 176 |
def run_wav2lip(video_path: str, audio_path: str) -> str:
|
| 177 |
"""Run Wav2Lip inference and return path to output video."""
|
| 178 |
-
out_path = str(OUTPUT_DIR / f"
|
| 179 |
tmp_out = str(WAV2LIP_DIR / "results" / "result_voice.mp4")
|
| 180 |
os.makedirs(os.path.dirname(tmp_out), exist_ok=True)
|
| 181 |
|
|
@@ -186,7 +158,6 @@ def run_wav2lip(video_path: str, audio_path: str) -> str:
|
|
| 186 |
"--face", video_path,
|
| 187 |
"--audio", audio_path,
|
| 188 |
"--outfile", tmp_out,
|
| 189 |
-
"--resize_factor", "2",
|
| 190 |
"--nosmooth",
|
| 191 |
]
|
| 192 |
print(f"[wav2lip] Running inference...")
|
|
@@ -201,19 +172,8 @@ def run_wav2lip(video_path: str, audio_path: str) -> str:
|
|
| 201 |
|
| 202 |
# ββ Main Pipeline βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 203 |
|
| 204 |
-
def
|
| 205 |
"""Main VoiceKit pipeline called by Gradio."""
|
| 206 |
-
global MODELS_READY, SETUP_OK, SETUP_ERR
|
| 207 |
-
|
| 208 |
-
if not MODELS_READY:
|
| 209 |
-
progress(0.01, desc="Downloading Wav2Lip models & checking configurations (first-time setup)...")
|
| 210 |
-
try:
|
| 211 |
-
bootstrap()
|
| 212 |
-
MODELS_READY = True
|
| 213 |
-
except Exception as e:
|
| 214 |
-
SETUP_OK = False
|
| 215 |
-
SETUP_ERR = str(e)
|
| 216 |
-
raise gr.Error(f"Setup failed: {e}")
|
| 217 |
|
| 218 |
if not SETUP_OK:
|
| 219 |
raise gr.Error(f"Setup failed: {SETUP_ERR}")
|
|
@@ -226,17 +186,17 @@ def voicekit_generate(video_file, voice_file, script, use_voice_clone, accent, p
|
|
| 226 |
|
| 227 |
# Step 1: Generate audio
|
| 228 |
if use_voice_clone and voice_file is not None:
|
| 229 |
-
progress(0.15, desc="
|
| 230 |
try:
|
| 231 |
audio_path = generate_voice(script, voice_file)
|
| 232 |
-
voice_note = "Voice
|
| 233 |
except Exception as e:
|
| 234 |
-
progress(0.15, desc=f"Voice
|
| 235 |
-
audio_path = generate_voice_edgetts(script
|
| 236 |
-
voice_note = f"Used generic voice (
|
| 237 |
else:
|
| 238 |
progress(0.2, desc="Generating audio with generic voice...")
|
| 239 |
-
audio_path = generate_voice_edgetts(script
|
| 240 |
voice_note = "Used generic voice (no voice sample provided)"
|
| 241 |
|
| 242 |
progress(0.5, desc="Running Wav2Lip lip-sync (1-5 min on CPU, ~1 min on GPU)...")
|
|
@@ -275,7 +235,11 @@ INSTRUCTIONS = """
|
|
| 275 |
- The face should be **clearly visible** throughout the video
|
| 276 |
"""
|
| 277 |
|
| 278 |
-
STATUS_MSG =
|
|
|
|
|
|
|
|
|
|
|
|
|
| 279 |
|
| 280 |
css = """
|
| 281 |
body { font-family: 'Inter', sans-serif; }
|
|
@@ -298,9 +262,9 @@ with gr.Blocks(
|
|
| 298 |
with gr.Row():
|
| 299 |
with gr.Column(scale=3):
|
| 300 |
gr.Markdown("### Step 1 β Upload Your Face Video")
|
| 301 |
-
video_input = gr.
|
| 302 |
label="Face Video (MP4 / MKV)",
|
| 303 |
-
|
| 304 |
)
|
| 305 |
|
| 306 |
gr.Markdown("### Step 2 β Upload Your Voice Sample")
|
|
@@ -314,19 +278,13 @@ with gr.Blocks(
|
|
| 314 |
value=True,
|
| 315 |
)
|
| 316 |
|
| 317 |
-
gr.Markdown("### Step 3 β Type Your Script
|
| 318 |
script_input = gr.Textbox(
|
| 319 |
label="Script",
|
| 320 |
placeholder="Hey [Name], I've been watching what you're doing and honestly there's a way to 10x your outreach using AI. Worth 2 minutes of your time?",
|
| 321 |
lines=6,
|
| 322 |
max_lines=20,
|
| 323 |
)
|
| 324 |
-
|
| 325 |
-
accent_input = gr.Radio(
|
| 326 |
-
["us", "uk"],
|
| 327 |
-
label="Accent (used for generic voice)",
|
| 328 |
-
value="us"
|
| 329 |
-
)
|
| 330 |
|
| 331 |
generate_btn = gr.Button(
|
| 332 |
"Generate Lip-Synced Video",
|
|
@@ -336,7 +294,7 @@ with gr.Blocks(
|
|
| 336 |
|
| 337 |
with gr.Column(scale=2):
|
| 338 |
gr.Markdown("### Output")
|
| 339 |
-
video_output = gr.
|
| 340 |
status_text = gr.Textbox(label="Status", interactive=False)
|
| 341 |
|
| 342 |
gr.Markdown(INSTRUCTIONS)
|
|
@@ -347,18 +305,17 @@ with gr.Blocks(
|
|
| 347 |
)
|
| 348 |
|
| 349 |
generate_btn.click(
|
| 350 |
-
fn=
|
| 351 |
-
inputs=[video_input, voice_input, script_input, use_clone
|
| 352 |
outputs=[video_output, status_text],
|
| 353 |
-
api_name="lipkit_generate",
|
| 354 |
)
|
| 355 |
|
| 356 |
gr.Examples(
|
| 357 |
examples=[
|
| 358 |
-
[None, None, "Hey, I've been keeping an eye on your content and honestly β the engagement you're getting is impressive. I built an AI tool that could help you turn one video into 50 personalized ones automatically. Want to see how it works?"
|
| 359 |
-
[None, None, "What's up! I help fintech founders like you turn one video into a hundred personalized outreach clips using AI. No editing skills needed, no extra time. Want me to show you a quick demo?"
|
| 360 |
],
|
| 361 |
-
inputs=[video_input, voice_input, script_input
|
| 362 |
label="Script Examples",
|
| 363 |
)
|
| 364 |
|
|
|
|
| 16 |
import urllib.request
|
| 17 |
from pathlib import Path
|
| 18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
import gradio as gr
|
| 20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
# ββ Paths βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 22 |
MODELS_DIR = Path("/tmp/voicekit_models")
|
| 23 |
WAV2LIP_DIR = MODELS_DIR / "Wav2Lip"
|
|
|
|
| 32 |
"https://huggingface.co/numz/wav2lip_studio/resolve/main/Wav2lip/wav2lip_gan.pth",
|
| 33 |
"https://huggingface.co/camenduru/wav2lip/resolve/main/wav2lip_gan.pth",
|
| 34 |
]
|
| 35 |
+
FACE_DET_URL = "https://huggingface.co/camenduru/wav2lip/resolve/main/face_detection/s3fd-619a316812.pth"
|
| 36 |
|
| 37 |
|
| 38 |
# ββ One-time Model Bootstrap ββββββββββββββββββββββββββββββββββββββββββββββββββ
|
|
|
|
| 83 |
# Face detector
|
| 84 |
_download(FACE_DET_URL, FACE_DET, "S3FD face detector (~85MB)")
|
| 85 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
print("[ready] All models ready")
|
| 87 |
|
| 88 |
|
| 89 |
+
# Run bootstrap at startup
|
| 90 |
+
try:
|
| 91 |
+
bootstrap()
|
| 92 |
+
SETUP_OK = True
|
| 93 |
+
SETUP_ERR = ""
|
| 94 |
+
except Exception as e:
|
| 95 |
+
SETUP_OK = False
|
| 96 |
+
SETUP_ERR = str(e)
|
| 97 |
+
print(f"[ERROR] Bootstrap failed: {e}")
|
| 98 |
|
| 99 |
|
| 100 |
# ββ Voice Generation ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 101 |
|
| 102 |
def generate_voice(script: str, voice_sample_path: str) -> str:
|
| 103 |
+
"""Generate speech using Kokoro TTS (Python 3.14 compatible)."""
|
| 104 |
+
import numpy as np
|
| 105 |
+
import soundfile as sf
|
| 106 |
|
| 107 |
out_path = str(OUTPUT_DIR / f"voice_{int(time.time())}.wav")
|
| 108 |
+
|
|
|
|
| 109 |
try:
|
| 110 |
+
from kokoro_onnx import Kokoro
|
| 111 |
+
print("[tts] Using Kokoro ONNX")
|
| 112 |
+
kokoro = Kokoro("kokoro-v1.0.onnx", "voices-v1.0.bin")
|
| 113 |
+
samples, sr = kokoro.create(script, voice="af_heart", speed=1.0, lang="en-us")
|
| 114 |
+
sf.write(out_path, samples, sr)
|
| 115 |
+
return out_path
|
| 116 |
+
except Exception:
|
| 117 |
+
pass
|
| 118 |
+
|
| 119 |
+
try:
|
| 120 |
+
import kokoro as k
|
| 121 |
+
print("[tts] Using Kokoro pipeline")
|
| 122 |
+
pipeline = k.KPipeline(lang_code="en")
|
| 123 |
+
parts = [audio for _, _, audio in pipeline(script, voice="af_heart", speed=1.0)]
|
| 124 |
+
combined = np.concatenate(parts) if len(parts) > 1 else parts[0]
|
| 125 |
+
sf.write(out_path, combined, 24000)
|
| 126 |
return out_path
|
| 127 |
except Exception as e:
|
| 128 |
+
print(f"[tts] Kokoro failed: {e}, using edge-tts")
|
| 129 |
+
return generate_voice_edgetts(script)
|
| 130 |
|
| 131 |
|
| 132 |
+
def generate_voice_edgetts(script: str) -> str:
|
| 133 |
"""Fallback: edge-tts generic voice."""
|
| 134 |
import asyncio
|
| 135 |
import edge_tts
|
| 136 |
out_path = str(OUTPUT_DIR / f"voice_generic_{int(time.time())}.mp3")
|
| 137 |
|
|
|
|
|
|
|
| 138 |
async def _gen():
|
| 139 |
+
c = edge_tts.Communicate(script, "en-US-GuyNeural")
|
| 140 |
await c.save(out_path)
|
| 141 |
|
| 142 |
asyncio.run(_gen())
|
|
|
|
| 147 |
|
| 148 |
def run_wav2lip(video_path: str, audio_path: str) -> str:
|
| 149 |
"""Run Wav2Lip inference and return path to output video."""
|
| 150 |
+
out_path = str(OUTPUT_DIR / f"lipkit_{int(time.time())}.mp4")
|
| 151 |
tmp_out = str(WAV2LIP_DIR / "results" / "result_voice.mp4")
|
| 152 |
os.makedirs(os.path.dirname(tmp_out), exist_ok=True)
|
| 153 |
|
|
|
|
| 158 |
"--face", video_path,
|
| 159 |
"--audio", audio_path,
|
| 160 |
"--outfile", tmp_out,
|
|
|
|
| 161 |
"--nosmooth",
|
| 162 |
]
|
| 163 |
print(f"[wav2lip] Running inference...")
|
|
|
|
| 172 |
|
| 173 |
# ββ Main Pipeline βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 174 |
|
| 175 |
+
def lipkit_generate(video_file, voice_file, script, use_voice_clone, progress=gr.Progress()):
|
| 176 |
"""Main VoiceKit pipeline called by Gradio."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 177 |
|
| 178 |
if not SETUP_OK:
|
| 179 |
raise gr.Error(f"Setup failed: {SETUP_ERR}")
|
|
|
|
| 186 |
|
| 187 |
# Step 1: Generate audio
|
| 188 |
if use_voice_clone and voice_file is not None:
|
| 189 |
+
progress(0.15, desc="Generating audio with Kokoro...")
|
| 190 |
try:
|
| 191 |
audio_path = generate_voice(script, voice_file)
|
| 192 |
+
voice_note = "Voice generated successfully"
|
| 193 |
except Exception as e:
|
| 194 |
+
progress(0.15, desc=f"Voice generation failed ({e}), using generic voice...")
|
| 195 |
+
audio_path = generate_voice_edgetts(script)
|
| 196 |
+
voice_note = f"Used generic voice (generation failed: {e})"
|
| 197 |
else:
|
| 198 |
progress(0.2, desc="Generating audio with generic voice...")
|
| 199 |
+
audio_path = generate_voice_edgetts(script)
|
| 200 |
voice_note = "Used generic voice (no voice sample provided)"
|
| 201 |
|
| 202 |
progress(0.5, desc="Running Wav2Lip lip-sync (1-5 min on CPU, ~1 min on GPU)...")
|
|
|
|
| 235 |
- The face should be **clearly visible** throughout the video
|
| 236 |
"""
|
| 237 |
|
| 238 |
+
STATUS_MSG = (
|
| 239 |
+
"Models ready. ZeroGPU enabled for fast processing."
|
| 240 |
+
if SETUP_OK
|
| 241 |
+
else f"Setup error: {SETUP_ERR}"
|
| 242 |
+
)
|
| 243 |
|
| 244 |
css = """
|
| 245 |
body { font-family: 'Inter', sans-serif; }
|
|
|
|
| 262 |
with gr.Row():
|
| 263 |
with gr.Column(scale=3):
|
| 264 |
gr.Markdown("### Step 1 β Upload Your Face Video")
|
| 265 |
+
video_input = gr.Video(
|
| 266 |
label="Face Video (MP4 / MKV)",
|
| 267 |
+
sources=["upload"],
|
| 268 |
)
|
| 269 |
|
| 270 |
gr.Markdown("### Step 2 β Upload Your Voice Sample")
|
|
|
|
| 278 |
value=True,
|
| 279 |
)
|
| 280 |
|
| 281 |
+
gr.Markdown("### Step 3 β Type Your Script")
|
| 282 |
script_input = gr.Textbox(
|
| 283 |
label="Script",
|
| 284 |
placeholder="Hey [Name], I've been watching what you're doing and honestly there's a way to 10x your outreach using AI. Worth 2 minutes of your time?",
|
| 285 |
lines=6,
|
| 286 |
max_lines=20,
|
| 287 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 288 |
|
| 289 |
generate_btn = gr.Button(
|
| 290 |
"Generate Lip-Synced Video",
|
|
|
|
| 294 |
|
| 295 |
with gr.Column(scale=2):
|
| 296 |
gr.Markdown("### Output")
|
| 297 |
+
video_output = gr.Video(label="Your Lip-Synced Video")
|
| 298 |
status_text = gr.Textbox(label="Status", interactive=False)
|
| 299 |
|
| 300 |
gr.Markdown(INSTRUCTIONS)
|
|
|
|
| 305 |
)
|
| 306 |
|
| 307 |
generate_btn.click(
|
| 308 |
+
fn=lipkit_generate,
|
| 309 |
+
inputs=[video_input, voice_input, script_input, use_clone],
|
| 310 |
outputs=[video_output, status_text],
|
|
|
|
| 311 |
)
|
| 312 |
|
| 313 |
gr.Examples(
|
| 314 |
examples=[
|
| 315 |
+
[None, None, "Hey, I've been keeping an eye on your content and honestly β the engagement you're getting is impressive. I built an AI tool that could help you turn one video into 50 personalized ones automatically. Want to see how it works?"],
|
| 316 |
+
[None, None, "What's up! I help fintech founders like you turn one video into a hundred personalized outreach clips using AI. No editing skills needed, no extra time. Want me to show you a quick demo?"],
|
| 317 |
],
|
| 318 |
+
inputs=[video_input, voice_input, script_input],
|
| 319 |
label="Script Examples",
|
| 320 |
)
|
| 321 |
|