Update app.py

app.py

@@ -22,7 +22,9 @@ SYSTEM_INSTRUCTION = (
     "sensory information not present in the media."
 )
 
-# Helpers
+IMAGE_EXTS = (".jpg", ".jpeg", ".png", ".webp", ".gif")
+VIDEO_EXTS = (".mp4", ".mov", ".webm", ".mkv")
+
 def get_client(key: str = None):
     api_key = (key or "").strip() or DEFAULT_KEY
     return Mistral(api_key=api_key)
@@ -30,12 +32,16 @@ def get_client(key: str = None):
 def is_remote(src: str) -> bool:
     return bool(src) and src.startswith(("http://", "https://"))
 
+def ext_from_src(src: str) -> str:
+    _, ext = os.path.splitext((src or "").split("?")[0])
+    return ext.lower()
+
 def fetch_bytes(src: str, stream_threshold=20 * 1024 * 1024) -> bytes:
     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:
+            cl = r.headers.get("content-length")
+            if cl and int(cl) > stream_threshold:
                 fd, path = tempfile.mkstemp()
                 os.close(fd)
                 with open(path, "wb") as f:
@@ -49,8 +55,7 @@ def fetch_bytes(src: str, stream_threshold=20 * 1024 * 1024) -> bytes:
                 except Exception:
                     pass
                 return data
-            else:
-                return r.content
+            return r.content
     with open(src, "rb") as f:
         return f.read()
 
@@ -79,22 +84,13 @@ def save_bytes_to_temp(b: bytes, suffix: str):
         f.write(b)
     return path
 
-IMAGE_EXTS = (".jpg", ".jpeg", ".png", ".webp", ".gif")
-VIDEO_EXTS = (".mp4", ".mov", ".webm", ".mkv")
-
-def ext_from_src(src: str) -> str:
-    _, ext = os.path.splitext((src or "").split("?")[0])
-    return ext.lower()
-
 def choose_model_for_src(src: str):
     ext = ext_from_src(src)
     if ext in VIDEO_EXTS:
         return DEFAULT_VIDEO_MODEL
     if ext in IMAGE_EXTS:
         return DEFAULT_IMAGE_MODEL
-    if is_remote(src):
-        return DEFAULT_VIDEO_MODEL
-    return DEFAULT_IMAGE_MODEL
+    return DEFAULT_VIDEO_MODEL if is_remote(src) else DEFAULT_IMAGE_MODEL
 
 def build_messages_for_image(prompt: str, b64_jpg: str):
     return [
@@ -114,11 +110,20 @@ def build_messages_for_text(prompt: str, extra_text: str):
 def extract_delta(chunk):
     if not chunk:
         return None
+    # chunk.data.choices[0].delta.content is the typical shape from Mistral streaming
     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:
+        # common streaming shape: data.choices[0].delta.content
+        content = data.choices[0].delta.content
+        if content is None:
+            return None
+        return str(content)
+    except Exception:
+        pass
+    try:
+        # fallback: delta may be dict-like
         c = data.choices[0].delta
         if isinstance(c, dict):
             txt = c.get("content") or c.get("text")
@@ -128,14 +133,15 @@ def extract_delta(chunk):
     except Exception:
         pass
     try:
+        # non-stream full message shape
         msg = data.choices[0].message
         if isinstance(msg, dict):
-            txt = msg.get("content")
+            content = msg.get("content")
         else:
-            txt = getattr(msg, "content", None)
-        if txt is None:
+            content = getattr(msg, "content", None)
+        if content is None:
             return None
-        return str(txt)
+        return str(content)
     except Exception:
         pass
     try:
@@ -143,7 +149,6 @@ def extract_delta(chunk):
     except Exception:
         return None
 
-# Core: produce final concatenated text
 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.")
@@ -153,6 +158,7 @@ def generate_final_text(src: str, custom_prompt: str, api_key: str):
 
     def stream_and_collect(model, messages):
         try:
+            # try streaming API
             stream_gen = None
             try:
                 stream_gen = client.chat.stream(model=model, messages=messages)
@@ -163,11 +169,12 @@ def generate_final_text(src: str, custom_prompt: str, api_key: str):
                     d = extract_delta(chunk)
                     if d is None:
                         continue
-                    # ignore whitespace-only pieces unless parts is empty and meaningful
+                    # drop pure-whitespace pieces unless result empty
                     if d.strip() == "" and parts:
                         continue
                     parts.append(d)
                 return
+            # fallback to non-streaming complete
             res = client.chat.complete(model=model, messages=messages, stream=False)
             try:
                 choices = getattr(res, "choices", None) or res.get("choices", [])
@@ -211,13 +218,13 @@ 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
+    # Remote video: send URL as text (avoid streaming non-text types)
     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: extract one frame with ffmpeg
+    # Local video: try extract frame with ffmpeg and send as image
     tmp_media = None
     try:
         media_bytes = fetch_bytes(src)