Update video processor max pixels

processing_colqwen3.py

@@ -169,6 +169,20 @@ class ColQwen3Processor(ProcessorMixin):
         instance.image_processor.max_pixels = max_num_visual_tokens * tile * tile
         instance.image_processor.size["longest_edge"] = instance.image_processor.max_pixels
 
+        video_patch_size = getattr(instance.video_processor, "patch_size", None)
+        video_merge_size = getattr(instance.video_processor, "merge_size", None) or getattr(
+            instance.video_processor, "spatial_merge_size", None
+        )
+        video_temporal_patch_size = getattr(instance.video_processor, "temporal_patch_size", None)
+        if video_patch_size is None or video_merge_size is None or video_temporal_patch_size is None:
+            raise ValueError(
+                "Qwen3VL video processor is missing `patch_size`, `merge_size`/`spatial_merge_size`, or `temporal_patch_size`."
+            )
+        video_tile = video_patch_size * video_merge_size
+        # Include temporal patching so the visual token cap applies across space and time.
+        instance.video_processor.max_pixels = max_num_visual_tokens * video_tile * video_tile * video_temporal_patch_size
+        instance.video_processor.size["longest_edge"] = instance.video_processor.max_pixels
+
         return instance
 
     def __call__(