Add Gemini OCR and fix VRC log world handling

src/ocr/ocr_actions.py

@@ -1,4 +1,5 @@
 import os
+import time
 from datetime import datetime
 
 os.environ["FLAGS_enable_pir_api"] = "0"
@@ -33,9 +34,9 @@ def getOcrEngine():
         log("INFO", "PaddleOCR initializing")
         config = loadOcrConfig()
         options = {
-            "lang": "japan",
+            "lang": config["lang"],
             "use_angle_cls": config["use_angle_cls"],
-            "show_log": False,
+            "show_log": config["show_log"],
         }
 
         try:
@@ -47,6 +48,27 @@ def getOcrEngine():
     return _ocr_engine
 
 
+_gemini_client = None
+
+
+def getGeminiClient(api_key):
+    global _gemini_client
+
+    if _gemini_client is not None:
+        return _gemini_client
+
+    if not api_key:
+        raise RuntimeError("Gemini API key is not set in secrets.toml [ai]")
+
+    try:
+        from google import genai
+    except Exception as e:
+        raise RuntimeError(f"Gemini SDK is missing detail={e}")
+
+    _gemini_client = genai.Client(api_key=api_key)
+    return _gemini_client
+
+
 def saveOcrText(image_path, text):
     return appendRuntimeLog(
         "OCR TEXT",
@@ -149,18 +171,69 @@ def runPaddleOcr(engine, image_path):
     return engine.ocr(str(image_path))
 
 
-def runOcrFromImage(image_path, show_result=True):
-    log("ACTION", f"PaddleOCR start image={image_path}")
+def runGeminiOcr(image_path):
+    config = loadOcrConfig()
+    model_name = config["model_name"] or "gemini-2.5-flash"
+    prompt = (
+        "Extract every readable text fragment from this image with maximum accuracy. "
+        "Do not summarize, explain, translate, or guess. "
+        "Return only the text exactly as it appears, preserving line breaks and order."
+    )
 
     try:
-        image_path = preprocessOcrImage(image_path)
-        engine = getOcrEngine()
-        result = runPaddleOcr(engine, image_path)
+        from google.genai import types
     except Exception as e:
-        log("ERROR", f"PaddleOCR failed detail={e}")
-        return None
+        raise RuntimeError(f"Gemini types import failed detail={e}")
+
+    client = getGeminiClient(config["api_key"])
+    mime_type = "image/png"
+    suffix = str(image_path).lower()
+    if suffix.endswith(".jpg") or suffix.endswith(".jpeg"):
+        mime_type = "image/jpeg"
+    elif suffix.endswith(".webp"):
+        mime_type = "image/webp"
+
+    image_bytes = image_path.read_bytes()
+    image_part = types.Part.from_bytes(data=image_bytes, mime_type=mime_type)
+    response = client.models.generate_content(
+        model=model_name,
+        contents=[prompt, image_part],
+    )
+    text = (response.text or "").strip()
+    return text
+
+
+def runOcrFromImage(image_path, show_result=True):
+    try:
+        image_path = preprocessOcrImage(image_path)
+        config = loadOcrConfig()
+        provider = config["provider"]
+        log("ACTION", f"OCR start provider={provider} image={image_path}")
+
+        if provider == "gemini":
+            text = runGeminiOcr(image_path)
+        else:
+            engine = getOcrEngine()
+            result = runPaddleOcr(engine, image_path)
+            text = extractTextFromAny(result)
+
+        if not text:
+            raise RuntimeError("OCR text is empty")
+    except Exception as e:
+        log("ERROR", f"OCR failed detail={e}")
+
+        if config.get("provider") == "gemini":
+            log("INFO", "Falling back to PaddleOCR")
+            try:
+                engine = getOcrEngine()
+                result = runPaddleOcr(engine, image_path)
+                text = extractTextFromAny(result)
+            except Exception as fallback_error:
+                log("ERROR", f"PaddleOCR fallback failed detail={fallback_error}")
+                return None
+        else:
+            return None
 
-    text = extractTextFromAny(result)
     text_path = saveOcrText(image_path, text)
 
     try:
@@ -183,4 +256,8 @@ def runOcrFromImage(image_path, show_result=True):
 def runOcrFromScreen():
     log("ACTION", "runOcrFromScreen called")
     image_path = captureOcrRegion()
+    delay_seconds = max(0.0, loadOcrConfig().get("delay_seconds", 1.0))
+    if delay_seconds:
+        log("INFO", f"OCR delayed seconds={delay_seconds}")
+        time.sleep(delay_seconds)
     return runOcrFromImage(image_path, show_result=True)