Fix image format to match OpenAI Responses API documentation

Co-authored-by: TomeHirata <[email protected]>

dspy/clients/lm.py

@@ -489,48 +489,15 @@ def _convert_content_item_to_responses_format(item: dict[str, Any]) -> dict[str,
     For images, converts from:
         {"type": "image_url", "image_url": {"url": "..."}}
     To:
-        {"type": "input_image", "source": {"type": "url", "url": "..."}}
-    or:
-        {"type": "input_image", "source": {"type": "base64", "media_type": "...", "data": "..."}}
+        {"type": "input_image", "image_url": "..."}
 
     For text and other types, passes through as-is (already in correct format).
     """
     if item.get("type") == "image_url":
         image_url = item.get("image_url", {}).get("url", "")
-
-        # Check if it's a base64 data URI
-        if image_url.startswith("data:"):
-            # Extract media type and base64 data
-            # Format: data:image/png;base64,iVBORw0KG...
-            parts = image_url.split(",", 1)
-            if len(parts) == 2:
-                header, data = parts
-                # Extract media type from header (e.g., "data:image/png;base64" -> "image/png")
-                # Handle both "data:image/png;base64" and "data:image/png" formats
-                media_type_parts = header.split(";")[0].replace("data:", "")
-                if media_type_parts:
-                    media_type = media_type_parts
-                else:
-                    # Fallback to a default media type if extraction fails
-                    media_type = "image/png"
-
-                return {
-                    "type": "input_image",
-                    "source": {
-                        "type": "base64",
-                        "media_type": media_type,
-                        "data": data,
-                    }
-                }
-            # If data URI is malformed (doesn't have comma separator), fall through to URL handling
-
-        # Otherwise treat as URL
         return {
             "type": "input_image",
-            "source": {
-                "type": "url",
-                "url": image_url,
-            }
+            "image_url": image_url,
         }
 
     # For non-image items, return as-is

tests/clients/test_lm.py

@@ -672,9 +672,7 @@ def test_responses_api_converts_images_correctly():
 
     # Second item should be converted to input_image format
     assert content[1]["type"] == "input_image"
-    assert content[1]["source"]["type"] == "base64"
-    assert content[1]["source"]["media_type"] == "image/png"
-    assert content[1]["source"]["data"] == "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg=="
+    assert content[1]["image_url"] == "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg=="
 
     # Test with URL image
     request_with_url_image = {
@@ -699,8 +697,7 @@ def test_responses_api_converts_images_correctly():
     content = result["input"][0]["content"]
     assert len(content) == 1
     assert content[0]["type"] == "input_image"
-    assert content[0]["source"]["type"] == "url"
-    assert content[0]["source"]["url"] == "https://example.com/image.jpg"
+    assert content[0]["image_url"] == "https://example.com/image.jpg"
 
 
 def test_responses_api_with_image_input():
@@ -759,5 +756,4 @@ def test_responses_api_with_image_input():
         # Check that image was converted to input_image format
         image_content = [c for c in content if c.get("type") == "input_image"]
         assert len(image_content) == 1
-        assert image_content[0]["source"]["type"] == "base64"
-        assert image_content[0]["source"]["media_type"] == "image/png"
+        assert image_content[0]["image_url"] == "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg=="