diff --git a/tests/models/multimodal/generation/vlm_utils/custom_inputs.py b/tests/models/multimodal/generation/vlm_utils/custom_inputs.py
index c68ac8f57662..e369416fc49c 100644
--- a/tests/models/multimodal/generation/vlm_utils/custom_inputs.py
+++ b/tests/models/multimodal/generation/vlm_utils/custom_inputs.py
@@ -1,12 +1,9 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Custom input builders for edge-cases in different models."""
-from io import BytesIO
 from typing import Callable
 
-import requests
-from PIL import Image
-
+from vllm.assets.image import ImageAsset
 from vllm.multimodal.image import rescale_image_size
 from vllm.multimodal.video import (rescale_video_size, resize_video,
                                    sample_frames_from_video)
@@ -118,9 +115,9 @@ def different_patch_input_cases_internvl():
 
 
 def windows_attention_image_qwen2_5_vl():
-    # image from regression issue: https://github.com/vllm-project/vllm/issues/15122
-    image_url = "https://github.com/AOMediaCodec/av1-avif/blob/main/testFiles/Link-U/hato.jpg?raw=true"
-    image = Image.open(BytesIO(requests.get(image_url).content))
+
+    # image from regression issue: https://github.com/vllm-project/vllm/issues/15122 # noqa: E501
+    image = ImageAsset("hato").pil_image
 
     question = "Describe the image."
     img_prompt = "<|vision_start|><|image_pad|><|vision_end|>"
diff --git a/vllm/assets/image.py b/vllm/assets/image.py
index c977242a3d48..c8f8d43a9835 100644
--- a/vllm/assets/image.py
+++ b/vllm/assets/image.py
@@ -11,7 +11,7 @@ from .base import get_vllm_public_assets
 
 VLM_IMAGES_DIR = "vision_model_images"
 
-ImageAssetName = Literal["stop_sign", "cherry_blossom"]
+ImageAssetName = Literal["stop_sign", "cherry_blossom", "hato"]
 
 
 @dataclass(frozen=True)