@@ -17,7 +17,7 @@
 
 from ...feature_extraction_utils import BatchFeature
 from ...image_utils import ImageInput, make_nested_list_of_images
-from ...processing_utils import ProcessingKwargs, ProcessorMixin, Unpack
+from ...processing_utils import MultiModalData, ProcessingKwargs, ProcessorMixin, Unpack
 from ...tokenization_utils_base import PreTokenizedInput, TextInput
 from ...utils import auto_docstring
 
@@ -98,9 +98,38 @@ def prepare_inputs_layout(self, images=None, text=None, audio=None, **kwargs):
                 text = [" ".join([self.image_token] * len(image_list)) for image_list in images]
 
         if audio is not None and not text:
-            text = " ".join([self.audio_token] * len(audio))
+            text = [self.audio_token] * len(audio)
         return images, text, None, audio
 
+    def _get_num_multimodal_tokens(self, image_sizes=None, audio_lengths=None, **kwargs):
+        """
+        Computes the number of placeholder tokens needed for multimodal inputs with the given sizes.
+
+        Args:
+            image_sizes (`list[list[int]]`, *optional*):
+                The input sizes formatted as (height, width) per each image.
+            audio_lengths (`list[int]`, *optional*):
+                The lengths of audio inputs in number of samples.
+
+        Returns:
+            `MultiModalData`: A `MultiModalData` object holding number of tokens per each of the provided
+            input modalities, along with other useful data.
+        """
+
+        vision_data = {}
+        if image_sizes is not None:
+            # NOTE: No image cropping supported yet
+            num_image_tokens = [self.image_seq_length] * len(image_sizes)
+            num_image_patches = [1] * len(image_sizes)
+
+            vision_data.update({"num_image_tokens": num_image_tokens, "num_image_patches": num_image_patches})
+
+        if audio_lengths is not None:
+            # NOTE: Every audio is padded or truncated to a fixed window
+            vision_data["num_audio_tokens"] = [self.audio_seq_length] * len(audio_lengths)
+
+        return MultiModalData(**vision_data)
+
     @property
     def model_input_names(self):
         tokenizer_input_names = self.tokenizer.model_input_names + ["token_type_ids"]
 
@@ -259,8 +259,8 @@ def _get_num_multimodal_tokens(self, image_sizes=None, video_sizes=None, audio_l
             input modalities, along with other useful data.
         """
 
-        images_kwargs = Gemma4ProcessorKwargs._defaults.get("images_kwargs", {})
-        images_kwargs.update(kwargs)
+        merged_kwargs = self._merge_kwargs(self.valid_processor_kwargs, **kwargs)
+        images_kwargs = merged_kwargs["images_kwargs"]
         patch_size = images_kwargs.get("patch_size", None) or self.image_processor.patch_size
         pooling_kernel_size = (
             images_kwargs.get("pooling_kernel_size", None) or self.image_processor.pooling_kernel_size
@@ -288,8 +288,7 @@ def _get_num_multimodal_tokens(self, image_sizes=None, video_sizes=None, audio_l
             vision_data.update({"num_image_tokens": num_image_tokens, "num_image_patches": num_image_patches})
 
         if video_sizes is not None:
-            videos_kwargs = Gemma4ProcessorKwargs._defaults.get("videos_kwargs", {})
-            videos_kwargs.update(kwargs)
+            videos_kwargs = merged_kwargs["videos_kwargs"]
             patch_size = videos_kwargs.get("patch_size", None) or self.video_processor.patch_size
             pooling_kernel_size = (
                 videos_kwargs.get("pooling_kernel_size", None) or self.video_processor.pooling_kernel_size