Update app.py

app.py

@@ -31,13 +31,11 @@ def is_remote(src: str) -> bool:
     return bool(src) and src.startswith(("http://", "https://"))
 
 def fetch_bytes(src: str, stream_threshold=20 * 1024 * 1024) -> bytes:
-    """Fetch bytes. For remote large files stream to a temp file then read."""
     if is_remote(src):
         with requests.get(src, timeout=60, stream=True) as r:
             r.raise_for_status()
             content_length = r.headers.get("content-length")
             if content_length and int(content_length) > stream_threshold:
-                # stream to temp file
                 fd, path = tempfile.mkstemp()
                 os.close(fd)
                 with open(path, "wb") as f:
@@ -59,7 +57,6 @@ def fetch_bytes(src: str, stream_threshold=20 * 1024 * 1024) -> bytes:
 def convert_to_jpeg_bytes(media_bytes: bytes, base_h=480) -> bytes:
     img = Image.open(BytesIO(media_bytes))
     try:
-        # take first frame for animated formats
         img.seek(0)
     except Exception:
         pass
@@ -95,7 +92,6 @@ def choose_model_for_src(src: str):
         return DEFAULT_VIDEO_MODEL
     if ext in IMAGE_EXTS:
         return DEFAULT_IMAGE_MODEL
-    # fallback: remote URLs likely videos, local files inspect existence
     if is_remote(src):
         return DEFAULT_VIDEO_MODEL
     return DEFAULT_IMAGE_MODEL
@@ -121,12 +117,14 @@ def extract_delta(chunk):
     data = getattr(chunk, "data", None) or getattr(chunk, "response", None) or getattr(chunk, "delta", None)
     if not data:
         return None
+    # Try common streaming shapes and coerce to trimmed string
     try:
-        # try common shapes and coerce to string
         c = data.choices[0].delta
         if isinstance(c, dict):
             txt = c.get("content") or c.get("text")
-            return str(txt) if txt is not None else None
+            if txt is None:
+                return None
+            return str(txt)
     except Exception:
         pass
     try:
@@ -135,7 +133,9 @@ def extract_delta(chunk):
             txt = msg.get("content")
         else:
             txt = getattr(msg, "content", None)
-        return str(txt) if txt is not None else None
+        if txt is None:
+            return None
+        return str(txt)
     except Exception:
         pass
     try:
@@ -147,14 +147,12 @@ def extract_delta(chunk):
 def generate_final_text(src: str, custom_prompt: str, api_key: str):
     client = get_client(api_key)
     prompt = (custom_prompt.strip() if custom_prompt and custom_prompt.strip() else "Please provide a detailed visual review.")
-    lower = (src or "").lower()
     ext = ext_from_src(src)
     is_image = ext in IMAGE_EXTS or (not is_remote(src) and os.path.isfile(src) and ext in IMAGE_EXTS)
     parts = []
 
     def stream_and_collect(model, messages):
         try:
-            # prefer streaming if available; fall back to non-streaming
             stream_gen = None
             try:
                 stream_gen = client.chat.stream(model=model, messages=messages)
@@ -163,12 +161,14 @@ def generate_final_text(src: str, custom_prompt: str, api_key: str):
             if stream_gen:
                 for chunk in stream_gen:
                     d = extract_delta(chunk)
-                    if d:
-                        parts.append(d)
+                    if d is None:
+                        continue
+                    # ignore whitespace-only pieces unless parts is empty and meaningful
+                    if d.strip() == "" and parts:
+                        continue
+                    parts.append(d)
                 return
-            # fallback complete
             res = client.chat.complete(model=model, messages=messages, stream=False)
-            # try to extract text
             try:
                 choices = getattr(res, "choices", None) or res.get("choices", [])
             except Exception:
@@ -184,7 +184,6 @@ def generate_final_text(src: str, custom_prompt: str, api_key: str):
                         if isinstance(content, str):
                             parts.append(content)
                         else:
-                            # handle list/dict shaped content
                             if isinstance(content, list):
                                 for c in content:
                                     if isinstance(c, dict) and c.get("type") == "text":
@@ -200,7 +199,7 @@ def generate_final_text(src: str, custom_prompt: str, api_key: str):
         except Exception as e:
             parts.append(f"[Model error: {e}]")
 
-    # Image path
+    # Image
     if is_image:
         try:
             raw = fetch_bytes(src)
@@ -212,13 +211,13 @@ def generate_final_text(src: str, custom_prompt: str, api_key: str):
         stream_and_collect(choose_model_for_src(src), msgs)
         return "".join(parts).strip()
 
-    # Remote video: send URL as plain text (avoid embedding non-text chunk types)
+    # Remote video
     if is_remote(src):
         msgs = build_messages_for_text(prompt, f"Video URL: {src}")
         stream_and_collect(choose_model_for_src(src), msgs)
         return "".join(parts).strip()
 
-    # Local video fallback: extract a frame with ffmpeg
+    # Local video: extract one frame with ffmpeg
     tmp_media = None
     try:
         media_bytes = fetch_bytes(src)
@@ -249,7 +248,6 @@ def generate_final_text(src: str, custom_prompt: str, api_key: str):
                         except Exception:
                             pass
                 else:
-                    # ffmpeg failed or produced no frame
                     try:
                         if tmp_frame and os.path.exists(tmp_frame):
                             os.remove(tmp_frame)
@@ -270,7 +268,7 @@ def generate_final_text(src: str, custom_prompt: str, api_key: str):
         except Exception:
             pass
 
-# Simple UI: left preview + inputs, right final text only (plain text)
+# UI
 css = """
 .preview_column { min-width: 380px; }
 .preview_media img, .preview_media video { max-width: 100%; height: auto; }