update gemini

app.py

@@ -128,16 +128,15 @@ def image_to_data_uri(image: Image.Image) -> str:
 # ============================================================================
 # STEP 1 — GEMINI 2.0 FLASH: GENERATE 5 DIVERSE CAPTIONS
 # Single API call — all 5 captions in one request
-# Avoids 429 rate limit that occurred with 5 separate calls
+# Retry logic: tries gemini-2.0-flash first, falls back to gemini-1.5-flash-8b
+# gemini-1.5-flash-8b has separate quota pool from gemini-2.0-flash
 # ============================================================================
 def generate_captions_gemini(image: Image.Image) -> list:
 
-    model = genai.GenerativeModel("gemini-2.0-flash")
-
     prompt = """Look at this image carefully and write 5 different captions from different perspectives.
 
-1. Overall scene:  describing the image in every possible detail in simple language.
-2. People: Describe the people, their clothing colors, style, and what they are doing.
+1. Overall scene: One sentence describing the general scene.
+2. People: Describe the people, their clothing colors, style, and what they are doing in detail.
 3. Background: Describe the background, setting, and surroundings.
 4. Objects: Describe the objects, plants, and items visible in the image.
 5. Full description: A complete description covering who is in the image, what they are doing, their appearance, and where the scene takes place.
@@ -149,26 +148,51 @@ CAPTION_3: [your caption here]
 CAPTION_4: [your caption here]
 CAPTION_5: [your caption here]"""
 
-    try:
-        response = model.generate_content([prompt, image])
-        raw_text = response.text.strip()
-
-        captions = []
-        for i in range(1, 6):
-            marker      = f"CAPTION_{i}:"
-            next_marker = f"CAPTION_{i+1}:" if i < 5 else None
-
-            if marker in raw_text:
-                start = raw_text.index(marker) + len(marker)
-                end   = raw_text.index(next_marker) if next_marker and next_marker in raw_text else len(raw_text)
-                cap   = raw_text[start:end].strip().lower()
-                captions.append(cap if cap else "a scene shown in the image")
-            else:
-                captions.append("a scene shown in the image")
+    # Try primary model first, fallback to secondary if quota exceeded
+    models_to_try = [
+        "gemini-2.0-flash",
+        "gemini-1.5-flash-8b",
+        "gemini-1.5-flash"
+    ]
 
-    except Exception as e:
-        st.warning(f"Gemini error: {str(e)[:80]}")
-        captions = ["a scene shown in the image"] * 5
+    raw_text = None
+
+    for model_name in models_to_try:
+        try:
+            model    = genai.GenerativeModel(model_name)
+            response = model.generate_content([prompt, image])
+            raw_text = response.text.strip()
+            break
+        except Exception as e:
+            error_msg = str(e)
+            if "429" in error_msg:
+                st.warning(f"{model_name} quota exceeded, trying next model...")
+                continue
+            else:
+                st.warning(f"Gemini error ({model_name}): {error_msg[:80]}")
+                continue
+
+    if raw_text is None:
+        st.error(
+            "All Gemini models hit quota limit. "
+            "Quota resets at midnight (Pacific Time). "
+            "Using fallback captions for now."
+        )
+        return ["a scene shown in the image"] * 5
+
+    # Parse the 5 captions from structured response
+    captions = []
+    for i in range(1, 6):
+        marker      = f"CAPTION_{i}:"
+        next_marker = f"CAPTION_{i+1}:" if i < 5 else None
+
+        if marker in raw_text:
+            start = raw_text.index(marker) + len(marker)
+            end   = raw_text.index(next_marker) if next_marker and next_marker in raw_text else len(raw_text)
+            cap   = raw_text[start:end].strip().lower()
+            captions.append(cap if cap else "a scene shown in the image")
+        else:
+            captions.append("a scene shown in the image")
 
     seen, unique = set(), []
     for c in captions: