diff --git a/anansyTypes/schemas.py b/anansyTypes/schemas.py
index 3f7b021..0b53f87 100644
--- a/anansyTypes/schemas.py
+++ b/anansyTypes/schemas.py
@@ -21,12 +21,6 @@ class ModifiedBaseModel(PydanticBaseModel):
                 'content': v.content,
                 'embedding': {'data': v.embedding.data.tolist()}
             }
-        if isinstance(v, Symbol):
-            return {
-                'id': v.id,
-                'label': v.label,
-                'image': v.image.tolist() if hasattr(v.image, 'tolist') else v.image
-            }
         if isinstance(v, Embedding):
             return {'data': v.data.tolist()}
         if isinstance(v, list):
@@ -56,7 +50,7 @@ class GetSymbolRequest(ModifiedBaseModel):
     lexeme_id: int
 
 class GetSymbolResponse(ModifiedBaseModel):
-    symbol: dict
+    symbol: Union[dict, None]
 
 class GetWordRequest(ModifiedBaseModel):
     word_text: str
diff --git a/anansyTypes/structures/symbol.py b/anansyTypes/structures/symbol.py
index 78dfc05..8af1f4b 100644
--- a/anansyTypes/structures/symbol.py
+++ b/anansyTypes/structures/symbol.py
@@ -4,6 +4,7 @@ import numpy as np
 from pydantic_tensor import Tensor
 from typing import Any, Literal
 from pydantic_core import core_schema
+import base64
 
 
 @dataclass
@@ -13,17 +14,25 @@ class Symbol:
     image: Tensor[np.ndarray[Any, Any], tuple, Literal["int32", "int64"]]
 
     @classmethod
-    def from_dict(cls, data: dict) -> 'Symbol':
+    def from_dict(cls, data: dict) -> "Symbol":
+        if isinstance(data["image"], str):
+            # Decode from base64
+            image_bytes = base64.b64decode(data["image"])
+            image_array = np.frombuffer(image_bytes, dtype=np.int32)
+        else:
+            # From list/array
+            image_array = np.array(data["image"], dtype=np.int32)
+
         return cls(
-            id=data['id'],
-            label=data['label'],
-            image=np.array(data['image'], dtype=np.int32)
+            id=data["id"],
+            label=data["label"],
+            image=image_array,
         )
 
     @classmethod
     def __get_pydantic_core_schema__(cls, source_type: Any, handler):
         from pydantic_core import core_schema
-        
+
         return core_schema.no_info_after_validator_function(
             cls,
             core_schema.dataclass_schema(
@@ -31,22 +40,34 @@ class Symbol:
                 schema=core_schema.dataclass_args_schema(
                     cls.__name__,
                     [
-                        core_schema.dataclass_field(name='id', schema=handler.generate_schema(int)),
-                        core_schema.dataclass_field(name='label', schema=handler.generate_schema(str)),
                         core_schema.dataclass_field(
-                            name='image',
-                            schema=handler.generate_schema(Tensor[np.ndarray[Any, Any], tuple, Literal["int32", "int64"]])
+                            name="id", schema=handler.generate_schema(int)
+                        ),
+                        core_schema.dataclass_field(
+                            name="label", schema=handler.generate_schema(str)
+                        ),
+                        core_schema.dataclass_field(
+                            name="image",
+                            schema=handler.generate_schema(
+                                Tensor[
+                                    np.ndarray[Any, Any],
+                                    tuple,
+                                    Literal["int32", "int64"],
+                                ]
+                            ),
                         ),
                     ],
                 ),
-                fields=['id', 'label', 'image'],
+                fields=["id", "label", "image"],
             ),
             serialization=core_schema.plain_serializer_function_ser_schema(
                 lambda x: {
-                    'id': x.id,
-                    'label': x.label,
-                    'image': x.image.tolist() if hasattr(x.image, 'tolist') else list(x.image)
+                    "id": x.id,
+                    "label": x.label,
+                    "image": base64.b64encode(np.array(x.image).tobytes()).decode(
+                        "utf-8"
+                    ),
                 },
-                when_used='json',
-            )
+                when_used="json",
+            ),
         )