diff --git a/deepseek_vl/utils/io.py b/deepseek_vl/utils/io.py
index 081f7a2..b71432a 100644
--- a/deepseek_vl/utils/io.py
+++ b/deepseek_vl/utils/io.py
@@ -22,6 +22,8 @@ from typing import Dict, List
 
 import PIL.Image
 import torch
+import base64
+import io
 from transformers import AutoModelForCausalLM
 
 from deepseek_vl.models import MultiModalityCausalLM, VLChatProcessor
@@ -42,6 +44,8 @@ def load_pretrained_model(model_path: str):
 def load_pil_images(conversations: List[Dict[str, str]]) -> List[PIL.Image.Image]:
     """
 
+    Support file path or base64 images.
+
     Args:
         conversations (List[Dict[str, str]]): the conversations with a list of messages. An example is :
             [
@@ -64,8 +68,15 @@ def load_pil_images(conversations: List[Dict[str, str]]) -> List[PIL.Image.Image
         if "images" not in message:
             continue
 
-        for image_path in message["images"]:
-            pil_img = PIL.Image.open(image_path)
+        for image_data in message["images"]:
+            if image_data.startswith("data:image"):
+                # Image data is in base64 format
+                _, image_data = image_data.split(",", 1)
+                image_bytes = base64.b64decode(image_data)
+                pil_img = PIL.Image.open(io.BytesIO(image_bytes))
+            else:
+                # Image data is a file path
+                pil_img = PIL.Image.open(image_data)
             pil_img = pil_img.convert("RGB")
             pil_images.append(pil_img)