Update streamlit_app.py

streamlit_app.py

@@ -7,6 +7,7 @@ import traceback
 from glob import glob
 from pathlib import Path
 from difflib import SequenceMatcher
+import concurrent.futures
 
 import yt_dlp
 import ffmpeg
@@ -169,7 +170,6 @@ default_prompt = (
 analysis_prompt = settings_exp.text_area("Enter analysis", value=default_prompt, height=140)
 settings_exp.text_input("Video Password (if needed)", key="video-password", placeholder="password", type="password")
 
-# Show which key is active
 key_source = "session" if st.session_state.get("api_key") else ".env" if os.getenv("GOOGLE_API_KEY") else "none"
 settings_exp.caption(f"Using API key from: **{key_source}**")
 
@@ -229,6 +229,137 @@ def remove_prompt_echo(prompt: str, text: str, check_len: int = 600, ratio_thres
             return b_full[len(ph):].lstrip(" \n:-")
     return text
 
+def compress_video_if_large(local_path: str, threshold_mb: int = 50):
+    try:
+        file_size_mb = os.path.getsize(local_path) / (1024 * 1024)
+    except Exception:
+        return local_path
+    if file_size_mb <= threshold_mb:
+        return local_path
+    compressed_path = str(Path(local_path).with_name(Path(local_path).stem + "_compressed.mp4"))
+    try:
+        return compress_video(local_path, compressed_path, crf=28, preset="fast")
+    except Exception:
+        return local_path
+
+def generate_via_responses_api(prompt_text: str, processed, model_used: str, max_tokens: int = 1024):
+    key = get_effective_api_key()
+    if not key:
+        raise RuntimeError("No API key provided")
+    if not HAS_GENAI or genai is None:
+        raise RuntimeError("Responses API not available; install google.generativeai SDK.")
+    genai.configure(api_key=key)
+    fname = file_name_or_id(processed)
+    if not fname:
+        raise RuntimeError("Uploaded file missing name/id")
+    system_msg = {"role": "system", "content": prompt_text}
+    user_msg = {"role": "user", "content": "Please summarize the attached video."}
+    try:
+        response = genai.responses.generate(
+            model=model_used,
+            messages=[system_msg, user_msg],
+            files=[{"name": fname}],
+            safety_settings=safety_settings,
+            max_output_tokens=max_tokens,
+        )
+    except TypeError:
+        response = genai.responses.generate(
+            model=model_used,
+            input=[{"text": prompt_text, "files": [{"name": fname}]}],
+            safety_settings=safety_settings,
+            max_output_tokens=max_tokens,
+        )
+
+    # Normalize outputs into text pieces
+    outputs = []
+    if response is None:
+        outputs = []
+    elif isinstance(response, dict):
+        for key in ("output", "candidates", "items", "responses"):
+            val = response.get(key)
+            if isinstance(val, list) and val:
+                outputs = val
+                break
+        if not outputs:
+            for v in response.values():
+                if isinstance(v, list) and v:
+                    outputs = v
+                    break
+    else:
+        for attr in ("output", "candidates", "items", "responses"):
+            val = getattr(response, attr, None)
+            if isinstance(val, list) and val:
+                outputs = val
+                break
+
+    if not isinstance(outputs, list):
+        outputs = list(outputs) if outputs else []
+
+    text_pieces = []
+    for item in outputs:
+        if item is None:
+            continue
+        cand_contents = None
+        if isinstance(item, dict):
+            for k in ("content", "text", "message", "output_text", "output"):
+                if k in item and item[k]:
+                    cand_contents = item[k]
+                    break
+        else:
+            for k in ("content", "text", "message", "output", "output_text"):
+                cand_contents = getattr(item, k, None)
+                if cand_contents:
+                    break
+
+        if isinstance(cand_contents, str):
+            if cand_contents.strip():
+                text_pieces.append(cand_contents.strip())
+            continue
+
+        if isinstance(cand_contents, (list, tuple)):
+            for c in cand_contents:
+                if c is None:
+                    continue
+                if isinstance(c, str):
+                    if c.strip():
+                        text_pieces.append(c.strip())
+                    continue
+                if isinstance(c, dict):
+                    t = c.get("text") or c.get("content")
+                else:
+                    t = getattr(c, "text", None) or getattr(c, "content", None)
+                if t:
+                    text_pieces.append(str(t).strip())
+            continue
+
+        direct = None
+        if isinstance(item, dict):
+            direct = item.get("text") or item.get("output_text") or item.get("message")
+        else:
+            direct = getattr(item, "text", None) or getattr(item, "output_text", None) or getattr(item, "message", None)
+        if direct:
+            text_pieces.append(str(direct).strip())
+
+    if not text_pieces:
+        top_text = None
+        if isinstance(response, dict):
+            top_text = response.get("text") or response.get("message")
+        else:
+            top_text = getattr(response, "text", None) or getattr(response, "message", None)
+        if top_text:
+            text_pieces.append(str(top_text).strip())
+
+    # Deduplicate, preserve order
+    seen = set()
+    filtered = []
+    for t in text_pieces:
+        if not isinstance(t, str):
+            continue
+        if t and t not in seen:
+            filtered.append(t)
+            seen.add(t)
+    return "\n\n".join(filtered)
+
 col1, col2 = st.columns([1, 3])
 with col1:
     generate_now = st.button("Generate the story", type="primary", disabled=not bool(get_effective_api_key()))
@@ -315,19 +446,7 @@ if generate_now and not st.session_state.get("busy"):
                     if not HAS_GENAI:
                         raise RuntimeError("google.generativeai SDK not available; install it.")
                     local_path = current_path
-                    upload_path = local_path
-                    try:
-                        file_size_mb = os.path.getsize(local_path) / (1024 * 1024)
-                    except Exception:
-                        file_size_mb = 0
-
-                    # If file is over 50MB, attempt compression. If compression fails, fall back to original file.
-                    if file_size_mb > 50:
-                        compressed_path = str(Path(local_path).with_name(Path(local_path).stem + "_compressed.mp4"))
-                        try:
-                            upload_path = compress_video(local_path, compressed_path, crf=28, preset="fast")
-                        except Exception:
-                            upload_path = local_path
+                    upload_path = compress_video_if_large(local_path)
 
                     with st.spinner("Uploading video..."):
                         uploaded = upload_video_sdk(upload_path)
@@ -338,15 +457,12 @@ if generate_now and not st.session_state.get("busy"):
                         st.session_state["file_hash"] = current_hash
 
                 prompt_text = (analysis_prompt.strip() or default_prompt).strip()
-
                 out = ""
                 model_used = model_id
                 max_tokens = 1024
-
                 est_tokens = max_tokens
-                est_cost_caption = f"Est. max tokens: {est_tokens}"
 
-                # First try Agent, but guard and FALLBACK to direct genai responses if Agent fails or returns empty.
+                # Try Agent first, fallback to Responses API
                 agent = maybe_create_agent(model_used)
                 debug_info = {"agent_attempted": False, "agent_ok": False, "agent_error": None, "agent_response_has_text": False}
                 if agent:
@@ -378,125 +494,14 @@ if generate_now and not st.session_state.get("busy"):
 
                 if not out:
                     try:
-                        if not HAS_GENAI or genai is None:
-                            raise RuntimeError("Responses API not available; install google.generativeai SDK.")
-                        genai.configure(api_key=key_to_use)
-                        fname = file_name_or_id(processed)
-                        if not fname:
-                            raise RuntimeError("Uploaded file missing name/id")
-                        system_msg = {"role": "system", "content": prompt_text}
-                        user_msg = {"role": "user", "content": "Please summarize the attached video."}
-
-                        try:
-                            response = genai.responses.generate(
-                                model=model_used,
-                                messages=[system_msg, user_msg],
-                                files=[{"name": fname}],
-                                safety_settings=safety_settings,
-                                max_output_tokens=max_tokens,
-                            )
-                        except TypeError:
-                            response = genai.responses.generate(
-                                model=model_used,
-                                input=[{"text": prompt_text, "files": [{"name": fname}]}],
-                                safety_settings=safety_settings,
-                                max_output_tokens=max_tokens,
-                            )
-
-                        outputs = []
-                        if response is None:
-                            outputs = []
-                        elif isinstance(response, dict):
-                            for key in ("output", "candidates", "items", "responses"):
-                                val = response.get(key)
-                                if isinstance(val, list) and val:
-                                    outputs = val
-                                    break
-                            if not outputs:
-                                for v in response.values():
-                                    if isinstance(v, list) and v:
-                                        outputs = v
-                                        break
-                        else:
-                            for attr in ("output", "candidates", "items", "responses"):
-                                val = getattr(response, attr, None)
-                                if isinstance(val, list) and val:
-                                    outputs = val
-                                    break
-
-                        if not isinstance(outputs, list):
-                            outputs = list(outputs) if outputs else []
-
-                        text_pieces = []
-                        for item in outputs:
-                            if item is None:
-                                continue
-                            cand_contents = None
-                            if isinstance(item, dict):
-                                for k in ("content", "text", "message", "output_text", "output"):
-                                    if k in item and item[k]:
-                                        cand_contents = item[k]
-                                        break
-                            else:
-                                for k in ("content", "text", "message", "output", "output_text"):
-                                    cand_contents = getattr(item, k, None)
-                                    if cand_contents:
-                                        break
-
-                            if isinstance(cand_contents, str):
-                                if cand_contents.strip():
-                                    text_pieces.append(cand_contents.strip())
-                                continue
-
-                            if isinstance(cand_contents, (list, tuple)):
-                                for c in cand_contents:
-                                    if c is None:
-                                        continue
-                                    if isinstance(c, str):
-                                        if c.strip():
-                                            text_pieces.append(c.strip())
-                                        continue
-                                    if isinstance(c, dict):
-                                        t = c.get("text") or c.get("content")
-                                    else:
-                                        t = getattr(c, "text", None) or getattr(c, "content", None)
-                                    if t:
-                                        text_pieces.append(str(t).strip())
-                                continue
-
-                            direct = None
-                            if isinstance(item, dict):
-                                direct = item.get("text") or item.get("output_text") or item.get("message")
-                            else:
-                                direct = getattr(item, "text", None) or getattr(item, "output_text", None) or getattr(item, "message", None)
-                            if direct:
-                                text_pieces.append(str(direct).strip())
-
-                        if not text_pieces:
-                            top_text = None
-                            if isinstance(response, dict):
-                                top_text = response.get("text") or response.get("message")
-                            else:
-                                top_text = getattr(response, "text", None) or getattr(response, "message", None)
-                            if top_text:
-                                text_pieces.append(str(top_text).strip())
-
-                        seen = set()
-                        filtered = []
-                        for t in text_pieces:
-                            if not isinstance(t, str):
-                                continue
-                            if t and t not in seen:
-                                filtered.append(t)
-                                seen.add(t)
-                        out = "\n\n".join(filtered)
+                        with st.spinner("Generating description via Responses API..."):
+                            out = generate_via_responses_api(prompt_text, processed, model_used, max_tokens=max_tokens)
                     except Exception as e:
                         tb = traceback.format_exc()
                         st.session_state["last_error"] = f"Responses API error: {e}\n\nDebug: {debug_info}\n\nTraceback:\n{tb}"
                         st.error("An error occurred while generating the story. You can try Generate again; the uploaded video will be reused.")
                         out = ""
 
-                # post-process output
                 if out:
                     out = remove_prompt_echo(prompt_text, out)
                     p = prompt_text