Skip to content

Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue)

Python

Set GPT 5.5 image detail to original

Abi Rajad74fc20375bd
diff --git a/backend/agent/providers/openai.py b/backend/agent/providers/openai.py
index 980db4d0..9a20613b 100644
--- a/backend/agent/providers/openai.py
+++ b/backend/agent/providers/openai.py
@@ -26,6 +26,7 @@ from llm import Llm, get_openai_api_name, get_openai_reasoning_effort
def _convert_message_to_responses_input(
message: ChatCompletionMessageParam,
+ image_detail: str = "high",
) -> Dict[str, Any]:
role = message.get("role", "user")
content = message.get("content", "")
@@ -46,13 +47,19 @@ def _convert_message_to_responses_input(
{
"type": "input_image",
"image_url": image_url.get("url", ""),
- "detail": image_url.get("detail", "high"),
+ "detail": image_detail,
}
)
return {"role": role, "content": parts}
+def _get_image_detail_for_model(model: Llm) -> str:
+ if get_openai_api_name(model) == "gpt-5.5":
+ return "original"
+ return "high"
+
+
def _get_event_attr(event: Any, key: str, default: Any = None) -> Any:
if hasattr(event, key):
return getattr(event, key)
@@ -420,8 +427,10 @@ class OpenAIProviderSession(ProviderSession):
model,
enabled=IS_DEBUG_ENABLED,
)
+ image_detail = _get_image_detail_for_model(model)
self._input_items: List[Dict[str, Any]] = [
- _convert_message_to_responses_input(message) for message in prompt_messages
+ _convert_message_to_responses_input(message, image_detail=image_detail)
+ for message in prompt_messages
]
async def stream_turn(self, on_event: EventSink) -> ProviderTurn:
diff --git a/backend/tests/test_openai_provider_session.py b/backend/tests/test_openai_provider_session.py
index fd88d5e0..991267ea 100644
--- a/backend/tests/test_openai_provider_session.py
+++ b/backend/tests/test_openai_provider_session.py
@@ -223,3 +223,69 @@ async def test_openai_provider_session_uses_gpt_5_5_high_reasoning_effort() -> N
assert first_call["model"] == "gpt-5.5"
assert first_call["reasoning"] == {"effort": "high", "summary": "auto"}
+
+
+async def test_openai_provider_session_uses_original_image_detail_for_gpt_5_5() -> None:
+ client = _FakeOpenAIClient()
+ session = OpenAIProviderSession(
+ client=client, # type: ignore[arg-type]
+ model=Llm.GPT_5_5_HIGH,
+ prompt_messages=[
+ {
+ "role": "user",
+ "content": [
+ {"type": "text", "text": "Build from this screenshot."},
+ {
+ "type": "image_url",
+ "image_url": {
+ "url": "data:image/png;base64,abc",
+ "detail": "high",
+ },
+ },
+ ],
+ }
+ ],
+ tools=_test_tools(),
+ )
+
+ await session.stream_turn(_noop_event_sink)
+
+ first_input = client.responses.calls[0]["input"]
+ image_part = first_input[0]["content"][1]
+
+ assert image_part["type"] == "input_image"
+ assert image_part["detail"] == "original"
+
+
+async def test_openai_provider_session_keeps_high_image_detail_for_non_gpt_5_5() -> None:
+ client = _FakeOpenAIClient()
+ session = OpenAIProviderSession(
+ client=client, # type: ignore[arg-type]
+ model=Llm.GPT_5_4_MINI_LOW,
+ prompt_messages=[
+ {
+ "role": "user",
+ "content": [
+ {"type": "text", "text": "Build from this screenshot."},
+ {
+ "type": "image_url",
+ "image_url": {
+ "url": "data:image/png;base64,abc",
+ "detail": "low",
+ },
+ },
+ ],
+ }
+ ],
+ tools=_test_tools(),
+ )
+
+ await session.stream_turn(_noop_event_sink)
+
+ first_input = client.responses.calls[0]["input"]
+ image_part = first_input[0]["content"][1]
+
+ assert image_part["type"] == "input_image"
+ assert image_part["detail"] == "high"
diff --git a/backend/tests/test_openai_reasoning_parser.py b/backend/tests/test_openai_reasoning_parser.py
index 61e7213b..129b6168 100644
--- a/backend/tests/test_openai_reasoning_parser.py
+++ b/backend/tests/test_openai_reasoning_parser.py
@@ -75,7 +75,7 @@ def test_convert_image_url_defaults_to_high_detail() -> None:
assert image_part["detail"] == "high"
-def test_convert_image_url_preserves_explicit_detail() -> None:
+def test_convert_image_url_uses_provider_detail_over_explicit_detail() -> None:
message = {
"role": "user",
"content": [
@@ -85,6 +85,6 @@ def test_convert_image_url_preserves_explicit_detail() -> None:
},
],
}
- result = _convert_message_to_responses_input(message) # type: ignore
+ result = _convert_message_to_responses_input(message, image_detail="original") # type: ignore
image_part = result["content"][0]
- assert image_part["detail"] == "low"
+ assert image_part["detail"] == "original"