fix: suppress ML library stdout noise in ocr.py and upscale.py

PaddleOCR prints download/init messages to stdout which corrupts the
JSON result that the bridge expects. Same risk with basicsr/realesrgan.
Applied the same fd-level stdout redirect pattern already used in
remove_bg.py: redirect fd 1 to stderr during ML work, restore for
the JSON result. Also added show_log=False to PaddleOCR constructor.

packages/ai/python/ocr.py

@@ -33,27 +33,39 @@ def run_tesseract(input_path, language):
 def run_paddleocr(input_path, language):
     """Run PaddleOCR."""
     os.environ["PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK"] = "True"
-    from paddleocr import PaddleOCR
-    from gpu import gpu_available
 
-    # Map API language codes to PaddleOCR codes
+    # Redirect stdout to stderr so PaddleOCR download/init messages
-    paddle_lang_map = {"en": "en", "de": "latin", "fr": "latin", "es": "latin", "zh": "ch", "ja": "japan", "ko": "korean"}
+    # cannot contaminate our JSON result on stdout.
-    paddle_lang = paddle_lang_map.get(language, "en")
+    stdout_fd = os.dup(1)
+    os.dup2(2, 1)
+
+    try:
+        from paddleocr import PaddleOCR
+        from gpu import gpu_available
+
+        # Map API language codes to PaddleOCR codes
+        paddle_lang_map = {"en": "en", "de": "latin", "fr": "latin", "es": "latin", "zh": "ch", "ja": "japan", "ko": "korean"}
+        paddle_lang = paddle_lang_map.get(language, "en")
+
+        emit_progress(20, "Loading")
+        ocr = PaddleOCR(lang=paddle_lang, use_gpu=gpu_available(), show_log=False)
+        emit_progress(30, "Scanning")
+        result = ocr.ocr(input_path)
+        emit_progress(70, "Extracting text")
+        text = "\n".join(
+            [
+                line[1][0]
+                for res in result
+                if res
+                for line in res
+                if line and line[1]
+            ]
+        )
+    finally:
+        # Restore stdout
+        os.dup2(stdout_fd, 1)
+        os.close(stdout_fd)
 
-    emit_progress(20, "Loading")
-    ocr = PaddleOCR(lang=paddle_lang, use_gpu=gpu_available())
-    emit_progress(30, "Scanning")
-    result = ocr.ocr(input_path)
-    emit_progress(70, "Extracting text")
-    text = "\n".join(
-        [
-            line[1][0]
-            for res in result
-            if res
-            for line in res
-            if line and line[1]
-        ]
-    )
     return text, "paddleocr"
 
 

packages/ai/python/upscale.py

@@ -31,11 +31,21 @@ def main():
 
         # Try Real-ESRGAN first
         try:
-            from basicsr.archs.rrdbnet_arch import RRDBNet
+            # Redirect stdout to stderr so basicsr/realesrgan init messages
-            from realesrgan import RealESRGANer
+            # cannot contaminate our JSON result on stdout.
-            from gpu import gpu_available
+            stdout_fd = os.dup(1)
-            import numpy as np
+            os.dup2(2, 1)
-            import torch
+
+            try:
+                from basicsr.archs.rrdbnet_arch import RRDBNet
+                from realesrgan import RealESRGANer
+                from gpu import gpu_available
+                import numpy as np
+                import torch
+            finally:
+                # Restore stdout after imports
+                os.dup2(stdout_fd, 1)
+                os.close(stdout_fd)
 
             if not os.path.exists(REALESRGAN_MODEL_PATH):
                 raise FileNotFoundError(f"RealESRGAN model not found: {REALESRGAN_MODEL_PATH}")