Change video downsampling logic to avoid CUDA OOM

app.py

@@ -62,7 +62,6 @@ def validate_media_constraints(message: dict, history: list[dict]) -> bool:
         if "<image>" in message["text"]:
             gr.Warning("Using <image> tags with video files is not supported.")
             return False
-        # TODO: Add frame count validation for videos similar to image count limits  # noqa: FIX002, TD002, TD003
     if video_count == 0 and image_count > MAX_NUM_IMAGES:
         gr.Warning(f"You can upload up to {MAX_NUM_IMAGES} images.")
         return False
@@ -77,10 +76,13 @@ def downsample_video(video_path: str) -> list[tuple[Image.Image, float]]:
     fps = vidcap.get(cv2.CAP_PROP_FPS)
     total_frames = int(vidcap.get(cv2.CAP_PROP_FRAME_COUNT))
 
-    frame_interval = int(fps / 3)
-    frames = []
+    frame_interval = max(total_frames // MAX_NUM_IMAGES, 1)
+    frames: list[tuple[Image.Image, float]] = []
+
+    for i in range(0, min(total_frames, MAX_NUM_IMAGES * frame_interval), frame_interval):
+        if len(frames) >= MAX_NUM_IMAGES:
+            break
 
-    for i in range(0, total_frames, frame_interval):
         vidcap.set(cv2.CAP_PROP_POS_FRAMES, i)
         success, image = vidcap.read()
         if success: