Update app.py

app.py

@@ -7,6 +7,7 @@ import subprocess
 import base64
 import json
 import shutil
+import tempfile
 from typing import Optional
 
 # 🎯 DÜZELTME 1: Google API için v1beta sürümünü zorunlu kılıyoruz. 
@@ -365,14 +366,15 @@ def run_production_pipeline(job: TranslationJob, gemini_key: str, elevenlabs_key
     if file_ext not in ['mp4', 'mov', 'webm']:
         file_ext = 'mp4'
         
-    # Temporary filenames
-    downloaded_video_path = f"temp_{video_id}_downloaded.{file_ext}"
-    input_video_path = f"temp_{video_id}_input.mp4"  # This will hold the standardized version
-    extracted_audio_path = f"temp_{video_id}_audio.mp3"
-    synthesized_audio_path = f"temp_{video_id}_tts.mp3"
-    srt_file_path = f"temp_{video_id}.srt"
-    output_video_path = f"temp_{video_id}_output.mp4"
-    lipsync_output_path = f"temp_{video_id}_lipsync.mp4"
+    # Use /tmp for temp files (reliable on Linux/HuggingFace Space, avoids CWD issues with FFmpeg)
+    tmp_dir = tempfile.gettempdir()
+    downloaded_video_path = os.path.join(tmp_dir, f"temp_{video_id}_downloaded.{file_ext}")
+    input_video_path = os.path.join(tmp_dir, f"temp_{video_id}_input.mp4")  # standardized version
+    extracted_audio_path = os.path.join(tmp_dir, f"temp_{video_id}_audio.mp3")
+    synthesized_audio_path = os.path.join(tmp_dir, f"temp_{video_id}_tts.mp3")
+    srt_file_path = os.path.join(tmp_dir, f"temp_{video_id}.srt")
+    output_video_path = os.path.join(tmp_dir, f"temp_{video_id}_output.mp4")
+    lipsync_output_path = os.path.join(tmp_dir, f"temp_{video_id}_lipsync.mp4")
     
     try:
         # 1. Download original video
@@ -559,9 +561,9 @@ def run_production_pipeline(job: TranslationJob, gemini_key: str, elevenlabs_key
             cmd_lipsync = [
                 "python", "inference.py",
                 "--checkpoint_path", "checkpoints/wav2lip_gan.pth",
-                "--face", f"../{input_video_path}",
-                "--audio", f"../{synthesized_audio_path}",
-                "--outfile", f"../{lipsync_output_path}"
+                "--face", input_video_path,
+                "--audio", synthesized_audio_path,
+                "--outfile", lipsync_output_path
             ]
             
             # Execute in the Wav2Lip directory to resolve relative imports
@@ -578,6 +580,11 @@ def run_production_pipeline(job: TranslationJob, gemini_key: str, elevenlabs_key
             
         if job.has_captions:
             print(f"[{video_id}] Burning subtitles with style '{job.caption_style}' using FFmpeg...")
+            
+            # Guard: verify SRT file was actually written before trying to burn it
+            if not os.path.exists(srt_file_path) or os.path.getsize(srt_file_path) == 0:
+                raise Exception(f"SRT subtitle file is missing or empty: {srt_file_path}")
+            
             style_str = get_ffmpeg_style(
                 job.caption_style,
                 custom_font=job.custom_font,
@@ -585,12 +592,23 @@ def run_production_pipeline(job: TranslationJob, gemini_key: str, elevenlabs_key
                 custom_color=job.custom_color
             )
             
+            # FFmpeg subtitles filter: path must use forward slashes.
+            # On Windows, the drive-letter colon (C:) must be escaped as C\:
+            # On Linux (HuggingFace Space), paths start with / so no escaping needed.
+            srt_filter_path = srt_file_path.replace("\\", "/")
+            if len(srt_filter_path) >= 2 and srt_filter_path[1] == ":":
+                # Windows drive letter colon escape
+                srt_filter_path = srt_filter_path[0] + "\\:" + srt_filter_path[2:]
+            
+            subtitles_filter = f"subtitles={srt_filter_path}:force_style='{style_str}'"
+            print(f"[{video_id}] FFmpeg subtitle filter: {subtitles_filter}")
+            
             if job.has_lip_sync:
                 # Lip-synced video already has the synthesized audio merged inside it.
                 # So we map audio track from the same file (0:a)
                 cmd_merge = [
                     "ffmpeg", "-y", "-i", video_src_for_subtitles,
-                    "-vf", f"subtitles={srt_file_path}:force_style='{style_str}'",
+                    "-vf", subtitles_filter,
                     "-map", "0:v", "-map", "0:a",
                     "-c:v", "libx264", "-pix_fmt", "yuv420p", "-c:a", "aac", "-shortest",
                     output_video_path
@@ -599,7 +617,7 @@ def run_production_pipeline(job: TranslationJob, gemini_key: str, elevenlabs_key
                 # Map audio from input 1 (synthesized audio path)
                 cmd_merge = [
                     "ffmpeg", "-y", "-i", video_src_for_subtitles, "-i", synthesized_audio_path,
-                    "-vf", f"subtitles={srt_file_path}:force_style='{style_str}'",
+                    "-vf", subtitles_filter,
                     "-map", "0:v", "-map", "1:a",
                     "-c:v", "libx264", "-pix_fmt", "yuv420p", "-c:a", "aac", "-shortest",
                     output_video_path