diff --git a/docs/web/integrations/embedding/voyageai.mdx b/docs/web/integrations/embedding/voyageai.mdx
index c0741b39b..957ab2b4d 100644
--- a/docs/web/integrations/embedding/voyageai.mdx
+++ b/docs/web/integrations/embedding/voyageai.mdx
@@ -57,7 +57,12 @@ Usage Example:
 
 ### Multimodal Example {#multimodal-example}
 
-The `voyage-multimodal-3.5` model can embed text alongside images. You can use image URLs, file paths, or PIL Image objects:
+The `voyage-multimodal-3.5` model can embed text alongside images. You can use image URLs, local `pathlib.Path` objects, raw image bytes, or PIL Image objects.
+
+For source embeddings, PIL images and images loaded from bytes or local paths are
+converted to RGB and encoded as JPEG. This supports RGBA, grayscale with alpha,
+and palette images; transparency is discarded. Composite transparent images onto
+your chosen background before embedding if the background color matters.
 
 <CodeGroup>
   <CodeBlock filename="Python" language="Python" icon="python">
diff --git a/python/python/lancedb/embeddings/voyageai.py b/python/python/lancedb/embeddings/voyageai.py
index 95525e07b..bcb2fd5a3 100644
--- a/python/python/lancedb/embeddings/voyageai.py
+++ b/python/python/lancedb/embeddings/voyageai.py
@@ -75,7 +75,7 @@ def transform_input(input_data: Union[str, bytes, Path]):
             content = {"type": "text", "text": input_data}
     elif isinstance(input_data, PIL_Image.Image):
         buffered = BytesIO()
-        input_data.save(buffered, format="JPEG")
+        input_data.convert("RGB").save(buffered, format="JPEG")
         img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
         content = {
             "type": "image_base64",
@@ -84,7 +84,7 @@ def transform_input(input_data: Union[str, bytes, Path]):
     elif isinstance(input_data, bytes):
         img = PIL_Image.open(BytesIO(input_data))
         buffered = BytesIO()
-        img.save(buffered, format="JPEG")
+        img.convert("RGB").save(buffered, format="JPEG")
         img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
         content = {
             "type": "image_base64",
@@ -103,7 +103,7 @@ def transform_input(input_data: Union[str, bytes, Path]):
         else:
             img = PIL_Image.open(input_data)
             buffered = BytesIO()
-            img.save(buffered, format="JPEG")
+            img.convert("RGB").save(buffered, format="JPEG")
             img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
             content = {
                 "type": "image_base64",
diff --git a/python/python/tests/test_voyageai_image_inputs.py b/python/python/tests/test_voyageai_image_inputs.py
new file mode 100644
index 000000000..885f94821
--- /dev/null
+++ b/python/python/tests/test_voyageai_image_inputs.py
@@ -0,0 +1,88 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+import base64
+from io import BytesIO
+from unittest.mock import MagicMock
+
+import pytest
+
+from lancedb.embeddings import get_registry
+from lancedb.embeddings.voyageai import VoyageAIEmbeddingFunction
+
+
+@pytest.mark.parametrize("model_name", ["voyage-multimodal-3", "voyage-multimodal-3.5"])
+@pytest.mark.parametrize("input_kind", ["pil", "bytes", "path"])
+@pytest.mark.parametrize(
+    "mode,color,expected_rgb",
+    [
+        ("RGBA", (64, 128, 192, 128), (64, 128, 192)),
+        ("LA", (96, 128), (96, 96, 96)),
+        ("P", 0, (64, 128, 192)),
+        ("RGB", (64, 128, 192), (64, 128, 192)),
+    ],
+)
+def test_multimodal_source_embeddings_encode_image_modes(
+    model_name, input_kind, mode, color, expected_rgb, tmp_path, monkeypatch
+):
+    Image = pytest.importorskip("PIL.Image")
+    image = Image.new(mode, (8, 6), color)
+    if mode == "P":
+        image.putpalette([64, 128, 192] + [0, 0, 0] * 255)
+        image.info["transparency"] = 0
+    original_pixels = image.tobytes()
+    image_input = image
+    if input_kind != "pil":
+        png = BytesIO()
+        image.save(png, format="PNG")
+        if input_kind == "bytes":
+            image_input = png.getvalue()
+        else:
+            image_input = tmp_path / "image.png"
+            image_input.write_bytes(png.getvalue())
+
+    client = MagicMock()
+    client.multimodal_embed.return_value.embeddings = [[0.1, 0.2]]
+    monkeypatch.setattr(VoyageAIEmbeddingFunction, "client", client)
+    func = get_registry().get("voyageai").create(name=model_name)
+
+    assert func.compute_source_embeddings(image_input) == [[0.1, 0.2]]
+
+    client.multimodal_embed.assert_called_once()
+    request = client.multimodal_embed.call_args.kwargs
+    assert request["model"] == model_name
+    assert request["input_type"] == "document"
+    assert len(request["inputs"]) == 1
+    (content,) = request["inputs"][0]["content"]
+    assert content["type"] == "image_base64"
+    prefix, encoded = content["image_base64"].split(",", 1)
+    assert prefix == "data:image/jpeg;base64"
+    with Image.open(BytesIO(base64.b64decode(encoded, validate=True))) as decoded:
+        assert decoded.format == "JPEG"
+        assert decoded.mode == "RGB"
+        assert decoded.size == (8, 6)
+        assert decoded.getpixel((0, 0)) == pytest.approx(expected_rgb, abs=3)
+    assert image.mode == mode
+    assert image.tobytes() == original_pixels
+    client.embed.assert_not_called()
+
+
+@pytest.mark.parametrize("input_kind", ["bytes", "path"])
+def test_multimodal_source_embeddings_reject_invalid_image(
+    input_kind, tmp_path, monkeypatch
+):
+    PIL = pytest.importorskip("PIL")
+    image_input = b"not an image"
+    if input_kind == "path":
+        image_input = tmp_path / "invalid.png"
+        image_input.write_bytes(b"not an image")
+
+    client = MagicMock()
+    monkeypatch.setattr(VoyageAIEmbeddingFunction, "client", client)
+    func = get_registry().get("voyageai").create(name="voyage-multimodal-3")
+
+    with pytest.raises(PIL.UnidentifiedImageError):
+        func.compute_source_embeddings(image_input)
+
+    client.multimodal_embed.assert_not_called()
+    client.embed.assert_not_called()