Skip to content

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

Python

Fix malformed Anthropic JSON

Abi Rajac15dd057d6e5
diff --git a/backend/routes/generate_code.py b/backend/routes/generate_code.py
index dfa511a6..8fc969ca 100644
--- a/backend/routes/generate_code.py
+++ b/backend/routes/generate_code.py
@@ -366,10 +366,11 @@ class ModelSelectionStage:
) -> List[Llm]:
"""Select appropriate models based on available API keys"""
try:
+ num_variants = 2 if generation_type == "update" else NUM_VARIANTS
variant_models = self._get_variant_models(
generation_type,
input_mode,
- NUM_VARIANTS,
+ num_variants,
openai_api_key,
anthropic_api_key,
gemini_api_key,
@@ -409,6 +410,14 @@ class ModelSelectionStage:
)
return list(VIDEO_VARIANT_MODELS)
+ # Edit/update mode prefers one Gemini + one OpenAI model for fast,
+ # complementary deltas.
+ if generation_type == "update" and gemini_api_key and openai_api_key:
+ return [
+ Llm.GEMINI_3_FLASH_PREVIEW_MINIMAL,
+ Llm.GPT_5_2_CODEX_LOW,
+ ]
+
# Define models based on available API keys
if gemini_api_key and anthropic_api_key and openai_api_key:
if input_mode == "text" and generation_type == "create":
@@ -684,10 +693,16 @@ class StatusBroadcastMiddleware(Middleware):
async def process(
self, context: PipelineContext, next_func: Callable[[], Awaitable[None]]
) -> None:
- # Determine variant count based on input mode
+ # Determine variant count based on input mode and generation type.
+ # Edit/update flows use two variants to keep latency and cost down.
assert context.extracted_params is not None
is_video_mode = context.extracted_params.input_mode == "video"
- num_variants = NUM_VARIANTS_VIDEO if is_video_mode else NUM_VARIANTS
+ is_update = context.extracted_params.generation_type == "update"
+ num_variants = (
+ NUM_VARIANTS_VIDEO
+ if is_video_mode
+ else 2 if is_update else NUM_VARIANTS
+ )
# Tell frontend how many variants we're using
await context.send_message("variantCount", str(num_variants), 0)
diff --git a/backend/routes/model_choice_sets.py b/backend/routes/model_choice_sets.py
index b1e4ffae..ac9b487b 100644
--- a/backend/routes/model_choice_sets.py
+++ b/backend/routes/model_choice_sets.py
@@ -15,8 +15,6 @@ ALL_KEYS_MODELS_TEXT_CREATE = (
)
ALL_KEYS_MODELS_UPDATE = (
Llm.GEMINI_3_FLASH_PREVIEW_MINIMAL,
- Llm.GPT_5_2_CODEX_HIGH,
- Llm.CLAUDE_SONNET_4_6,
Llm.GPT_5_2_CODEX_LOW,
)
ALL_KEYS_MODELS_DEFAULT = (
diff --git a/backend/tests/test_model_selection.py b/backend/tests/test_model_selection.py
index 72be4cd1..5dfbb73e 100644
--- a/backend/tests/test_model_selection.py
+++ b/backend/tests/test_model_selection.py
@@ -33,7 +33,7 @@ class TestModelSelectionAllKeys:
@pytest.mark.asyncio
async def test_gemini_anthropic_update_text(self):
- """All keys text update: uses edit-specific model mix."""
+ """All keys text update: uses two fast edit variants."""
models = await self.model_selector.select_models(
generation_type="update",
input_mode="text",
@@ -44,15 +44,13 @@ class TestModelSelectionAllKeys:
expected = [
Llm.GEMINI_3_FLASH_PREVIEW_MINIMAL,
- Llm.GPT_5_2_CODEX_HIGH,
- Llm.CLAUDE_SONNET_4_6,
Llm.GPT_5_2_CODEX_LOW,
]
assert models == expected
@pytest.mark.asyncio
async def test_gemini_anthropic_update(self):
- """All keys update: uses edit-specific model mix."""
+ """All keys image update: uses two fast edit variants."""
models = await self.model_selector.select_models(
generation_type="update",
input_mode="image",
@@ -63,8 +61,6 @@ class TestModelSelectionAllKeys:
expected = [
Llm.GEMINI_3_FLASH_PREVIEW_MINIMAL,
- Llm.GPT_5_2_CODEX_HIGH,
- Llm.CLAUDE_SONNET_4_6,
Llm.GPT_5_2_CODEX_LOW,
]
assert models == expected
diff --git a/backend/tests/test_status_broadcast.py b/backend/tests/test_status_broadcast.py
index 1024e4af..b5093513 100644
--- a/backend/tests/test_status_broadcast.py
+++ b/backend/tests/test_status_broadcast.py
@@ -60,3 +60,54 @@ async def test_video_update_broadcasts_two_variants() -> None:
assert len(status_messages) == 2
assert [m[2] for m in status_messages] == [0, 1]
assert next_called is True
+
+
+async def test_image_update_broadcasts_two_variants() -> None:
+ sent_messages: list[tuple[str, str | None, int]] = []
+
+ async def send_message(
+ msg_type: str,
+ value: str | None,
+ variant_index: int,
+ data=None,
+ eventId=None,
+ ) -> None:
+ sent_messages.append((msg_type, value, variant_index))
+
+ context = PipelineContext(websocket=MagicMock())
+ context.ws_comm = cast(
+ Any,
+ SimpleNamespace(
+ send_message=send_message,
+ throw_error=AsyncMock(),
+ ),
+ )
+ context.extracted_params = ExtractedParams(
+ stack="html_tailwind",
+ input_mode="image",
+ should_generate_images=True,
+ openai_api_key="key",
+ anthropic_api_key="key",
+ openai_base_url=None,
+ generation_type="update",
+ prompt={"text": "Edit this screenshot", "images": ["data:image/png;base64,abc"], "videos": []},
+ history=[],
+ file_state={"path": "index.html", "content": "<html></html>"},
+ option_codes=[],
+ )
+
+ middleware = StatusBroadcastMiddleware()
+ next_called = False
+
+ async def next_func() -> None:
+ nonlocal next_called
+ next_called = True
+
+ await middleware.process(context, next_func)
+
+ assert sent_messages[0] == ("variantCount", "2", 0)
+ status_messages = [m for m in sent_messages if m[0] == "status"]
+ assert len(status_messages) == 2
+ assert [m[2] for m in status_messages] == [0, 1]
+ assert next_called is True