diff --git a/core/grounding/fast_detector.py b/core/grounding/fast_detector.py
index 2a5984f60..63e39a2e5 100644
--- a/core/grounding/fast_detector.py
+++ b/core/grounding/fast_detector.py
@@ -141,18 +141,54 @@ class FastDetector:
     # OCR
     # ------------------------------------------------------------------
 
+    _easyocr_reader = None  # Singleton EasyOCR (chargé une fois)
+
     def _ocr_extract(self, image) -> List[Dict[str, Any]]:
-        """Extrait les mots visibles via docTR."""
+        """Extrait les mots visibles via EasyOCR (GPU, ~500ms).
+
+        Fallback sur docTR si EasyOCR non disponible.
+        """
         try:
-            import sys
-            sys.path.insert(0, 'visual_workflow_builder/backend')
-            from services.ocr_service import ocr_extract_words
+            import numpy as np
+            import easyocr
 
-            words = ocr_extract_words(image)
-            return words if words else []
+            # Singleton : charger le reader une seule fois
+            if FastDetector._easyocr_reader is None:
+                print(f"🔍 [FAST/ocr] Chargement EasyOCR (GPU)...")
+                FastDetector._easyocr_reader = easyocr.Reader(
+                    ['fr', 'en'], gpu=True, verbose=False
+                )
 
+            results = FastDetector._easyocr_reader.readtext(np.array(image))
+
+            words = []
+            for (bbox_pts, text, conf) in results:
+                if not text or len(text.strip()) < 1:
+                    continue
+                # bbox_pts = [[x1,y1],[x2,y1],[x2,y2],[x1,y2]]
+                x1 = int(min(p[0] for p in bbox_pts))
+                y1 = int(min(p[1] for p in bbox_pts))
+                x2 = int(max(p[0] for p in bbox_pts))
+                y2 = int(max(p[1] for p in bbox_pts))
+                words.append({
+                    'text': text.strip(),
+                    'bbox': [x1, y1, x2, y2],
+                    'confidence': float(conf),
+                })
+
+            return words
+
+        except ImportError:
+            # Fallback docTR
+            try:
+                import sys
+                sys.path.insert(0, 'visual_workflow_builder/backend')
+                from services.ocr_service import ocr_extract_words
+                return ocr_extract_words(image) or []
+            except Exception:
+                return []
         except Exception as ex:
-            print(f"⚠️ [FAST/ocr] docTR erreur: {ex}")
+            print(f"⚠️ [FAST/ocr] EasyOCR erreur: {ex}")
             return []
 
     # ------------------------------------------------------------------
diff --git a/core/grounding/title_verifier.py b/core/grounding/title_verifier.py
index dce14b828..3a87e7169 100644
--- a/core/grounding/title_verifier.py
+++ b/core/grounding/title_verifier.py
@@ -137,11 +137,33 @@ class TitleVerifier:
             'reason': 'Titre changé' if changed else 'Titre identique (acceptable)',
         }
 
+    _easyocr_reader = None  # Singleton partagé
+
     def _get_ocr(self):
-        """Lazy load de la fonction OCR."""
+        """Lazy load de la fonction OCR (EasyOCR prioritaire, fallback docTR)."""
         if self._ocr_fn is not None:
             return self._ocr_fn
 
+        # EasyOCR (rapide, bonne qualité GUI)
+        try:
+            import easyocr
+            import numpy as np
+
+            if TitleVerifier._easyocr_reader is None:
+                TitleVerifier._easyocr_reader = easyocr.Reader(
+                    ['fr', 'en'], gpu=True, verbose=False
+                )
+
+            def _easyocr_extract_text(img):
+                results = TitleVerifier._easyocr_reader.readtext(np.array(img))
+                return ' '.join(r[1] for r in results if r[1].strip())
+
+            self._ocr_fn = _easyocr_extract_text
+            return self._ocr_fn
+        except ImportError:
+            pass
+
+        # Fallback docTR
         try:
             import sys
             sys.path.insert(0, 'visual_workflow_builder/backend')
@@ -149,10 +171,4 @@ class TitleVerifier:
             self._ocr_fn = ocr_extract_text
             return self._ocr_fn
         except ImportError:
-            try:
-                from core.extraction.field_extractor import FieldExtractor
-                extractor = FieldExtractor()
-                self._ocr_fn = extractor.extract_text_from_image
-                return self._ocr_fn
-            except ImportError:
-                return None
+            return None