Update streamlit_app.py

streamlit_app.py

@@ -298,38 +298,69 @@ def get_runtime_api_key():
     return os.getenv("GOOGLE_API_KEY", "").strip() or None
 
 # Robust responses caller: SDK preferred, then attempt HTTP fallback via requests
+# --- REPLACE/ADD START ---
+import json
+import requests
+
 def responses_generate(model, messages, files, max_output_tokens, api_key):
     if not api_key:
         raise RuntimeError("No API key for responses_generate")
-    if not HAS_GENAI:
-        # try HTTP directly using requests
-        import requests
-        url = "https://gen-ai.googleapis.com/v1/responses"
-        payload = {"model": model, "messages": messages, "max_output_tokens": max_output_tokens}
-        headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
+    # If genai SDK present and has responses.generate, prefer it
+    if HAS_GENAI and genai is not None:
         try:
-            r = requests.post(url, json=payload, headers=headers, timeout=60)
-            r.raise_for_status()
-            return r.json()
+            genai.configure(api_key=api_key)
+            if hasattr(genai, "responses") and getattr(genai, "responses") is not None:
+                return genai.responses.generate(model=model, messages=messages, files=files, max_output_tokens=max_output_tokens)
         except Exception as e:
-            raise RuntimeError(f"HTTP fallback to Responses API failed: {e}")
-    genai.configure(api_key=api_key)
-    # preferred SDK call
-    if hasattr(genai, "responses") and getattr(genai, "responses") is not None:
-        return genai.responses.generate(model=model, messages=messages, files=files, max_output_tokens=max_output_tokens)
-    # try client-level post if exposed
-    client_attr = getattr(genai, "client", None) or getattr(genai, "_client", None) or getattr(genai, "transport", None)
-    if client_attr:
-        post = getattr(client_attr, "post", None) or getattr(client_attr, "request", None)
-        if post:
-            payload = {"model": model, "messages": messages, "max_output_tokens": max_output_tokens}
-            headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
-            try:
-                resp = post("/v1/responses", json=payload, headers=headers)
-                return resp
-            except Exception:
-                pass
-    raise RuntimeError("genai.responses not available and HTTP fallback failed")
+            # allow fallback to HTTP and surface SDK error in debug later
+            sdk_err = str(e)
+    else:
+        sdk_err = "SDK not available"
+
+    # HTTP fallback using requests with retries
+    url = "https://gen-ai.googleapis.com/v1/responses"
+    payload = {"model": model, "messages": messages, "max_output_tokens": max_output_tokens}
+    headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
+    last_exc = None
+    for attempt in range(3):
+        try:
+            resp = requests.post(url, json=payload, headers=headers, timeout=30)
+            # Try to parse JSON on success
+            if resp.status_code == 200:
+                try:
+                    return resp.json()
+                except Exception:
+                    return {"text": resp.text}
+            else:
+                last_exc = RuntimeError(f"HTTP {resp.status_code}: {resp.text}")
+                time.sleep(1 + attempt)
+        except Exception as e:
+            last_exc = e
+            time.sleep(1 + attempt)
+
+    # If we get here, both SDK and HTTP failed. Provide diagnostic info.
+    diag = {
+        "sdk_error": sdk_err if 'sdk_err' in locals() else None,
+        "http_error": str(last_exc),
+        "http_url": url,
+        "payload_sample": json.dumps(payload)[:1000],
+        "note": "See Streamlit 'Debug (compact)' output for response keys/body"
+    }
+    raise RuntimeError(f"genai.responses not available and HTTP fallback failed: {diag}")
+
+# call wrapper that returns response or raises; replace the previous call_responses_once with this:
+def call_responses_once(model_used, system_msg, user_msg, fname, max_tokens):
+    files = [{"name": fname}] if fname else None
+    for attempt in range(2):
+        try:
+            return responses_generate(model_used, [system_msg, user_msg], files, max_tokens, api_key=get_runtime_api_key())
+        except Exception as e:
+            if attempt == 0:
+                time.sleep(1.0)
+                continue
+            raise
+# --- REPLACE/ADD END ---
+
 
 if (st.session_state.get("busy") is False) and ('generate_now' in locals() and generate_now):
     if not st.session_state.get("videos"):