diff --git a/sleap_io/io/slp.py b/sleap_io/io/slp.py
index 4d31a577..1f172fe4 100644
--- a/sleap_io/io/slp.py
+++ b/sleap_io/io/slp.py
@@ -20,7 +20,7 @@
     LabeledFrame,
     Labels,
 )
-from sleap_io.io.video import MediaVideo, HDF5Video
+from sleap_io.io.video import ImageVideo, MediaVideo, HDF5Video
 from sleap_io.io.utils import (
     read_hdf5_attrs,
     read_hdf5_dataset,
@@ -72,16 +72,23 @@ def read_videos(labels_path: str) -> list[Video]:
                 # complex path finding strategies.
                 pass
 
+        video_path = video_path.as_posix()
+
+        if "filenames" in backend:
+            # This is an ImageVideo.
+            # TODO: Path resolution.
+            video_path = backend["filenames"]
+
         try:
             backend = VideoBackend.from_filename(
-                video_path.as_posix(),
+                video_path,
                 dataset=backend.get("dataset", None),
                 grayscale=backend.get("grayscale", None),
                 input_format=backend.get("input_format", None),
             )
         except ValueError:
             backend = None
-        video_objects.append(Video(filename=video_path.as_posix(), backend=backend))
+        video_objects.append(Video(filename=video_path, backend=backend))
     return video_objects
 
 
@@ -119,6 +126,24 @@ def write_videos(labels_path: str, videos: list[Video]):
             # TODO: Handle saving embedded images or restoring source video.
             # Ref: https://github.com/talmolab/sleap/blob/fb61b6ce7a9ac9613d99303111f3daafaffc299b/sleap/io/format/hdf5.py#L246-L273
 
+        elif type(video.backend) == ImageVideo:
+            shape = video.shape
+            if shape is None:
+                height, width, channels = 0, 0, 1
+            else:
+                height, width, channels = shape[1:]
+
+            video_json = {
+                "backend": {
+                    "filename": video.filename[0],
+                    "filenames": video.filename,
+                    "height_": height,
+                    "width_": width,
+                    "channels_": channels,
+                    "grayscale": video.backend.grayscale,
+                }
+            }
+
         else:
             raise NotImplementedError(
                 f"Cannot serialize video backend for video: {video}"
diff --git a/tests/data/slp/imgvideo.slp b/tests/data/slp/imgvideo.slp
new file mode 100644
index 00000000..0369499c
Binary files /dev/null and b/tests/data/slp/imgvideo.slp differ
diff --git a/tests/fixtures/slp.py b/tests/fixtures/slp.py
index 1b82a32e..b9ec1f4c 100644
--- a/tests/fixtures/slp.py
+++ b/tests/fixtures/slp.py
@@ -70,3 +70,9 @@ def slp_real_data():
         "/Users/talmo/sleap-io/tests/data/videos/centered_pair_low_quality.mp4"
     """
     return "tests/data/slp/labels.v002.rel_paths.slp"
+
+
+@pytest.fixture
+def slp_imgvideo():
+    """SLP project with a single image video."""
+    return "tests/data/slp/imgvideo.slp"
diff --git a/tests/io/test_slp.py b/tests/io/test_slp.py
index 399674c3..e2ab9470 100644
--- a/tests/io/test_slp.py
+++ b/tests/io/test_slp.py
@@ -33,6 +33,8 @@
 from sleap_io.io.utils import read_hdf5_dataset
 import numpy as np
 
+from sleap_io.io.video import ImageVideo
+
 
 def test_read_labels(slp_typical, slp_simple_skel, slp_minimal):
     """Test `read_labels` can read different types of .slp files."""
@@ -215,3 +217,23 @@ def test_load_multi_skeleton(tmpdir):
     assert loaded_skels[1].edge_inds == [(0, 1)]
     assert loaded_skels[0].flipped_node_inds == [1, 0]
     assert loaded_skels[1].flipped_node_inds == [1, 0]
+
+
+def test_slp_imgvideo(tmpdir, slp_imgvideo):
+    labels = read_labels(slp_imgvideo)
+    assert type(labels.video.backend) == ImageVideo
+    assert labels.video.shape == (3, 384, 384, 1)
+
+    write_labels(tmpdir / "test.slp", labels)
+    labels = read_labels(tmpdir / "test.slp")
+    assert type(labels.video.backend) == ImageVideo
+    assert labels.video.shape == (3, 384, 384, 1)
+
+    videos = [Video.from_filename(["fake1.jpg", "fake2.jpg"])]
+    assert videos[0].shape is None
+    assert len(videos[0].filename) == 2
+    write_videos(tmpdir / "test2.slp", videos)
+    videos = read_videos(tmpdir / "test2.slp")
+    assert type(videos[0].backend) == ImageVideo
+    assert len(videos[0].filename) == 2
+    assert videos[0].shape is None