mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-17 09:34:48 +00:00
fix(proxy): keep image-edit uploads as files instead of strings (#41421)
This commit is contained in:
parent
8fc9c46d1a
commit
4ba95ff4f2
3 changed files with 171 additions and 10 deletions
|
|
@ -75,7 +75,10 @@ class OpenAIImageEditConfig(BaseImageEditConfig):
|
|||
if isinstance(image, BufferedReader):
|
||||
files_list.append((field_name, (image.name, image, image_content_type)))
|
||||
else:
|
||||
files_list.append((field_name, ("image.png", image, image_content_type)))
|
||||
_filename = getattr(image, "name", None)
|
||||
if not isinstance(_filename, str) or not _filename:
|
||||
_filename = "image.png"
|
||||
files_list.append((field_name, (_filename, image, image_content_type)))
|
||||
|
||||
def transform_image_edit_request(
|
||||
self,
|
||||
|
|
@ -135,7 +138,10 @@ class OpenAIImageEditConfig(BaseImageEditConfig):
|
|||
if isinstance(_mask, BufferedReader):
|
||||
files_list.append(("mask", (_mask.name, _mask, mask_content_type)))
|
||||
else:
|
||||
files_list.append(("mask", ("mask.png", _mask, mask_content_type)))
|
||||
_mask_filename = getattr(_mask, "name", None)
|
||||
if not isinstance(_mask_filename, str) or not _mask_filename:
|
||||
_mask_filename = "mask.png"
|
||||
files_list.append(("mask", (_mask_filename, _mask, mask_content_type)))
|
||||
|
||||
return data_without_files, files_list
|
||||
|
||||
|
|
|
|||
|
|
@ -298,14 +298,18 @@ async def image_edit_api(
|
|||
#########################################################
|
||||
# Read request body and convert UploadFiles to BytesIO
|
||||
#########################################################
|
||||
data: Final = {
|
||||
key: value
|
||||
for key, value in coerce_numeric_form_fields(
|
||||
parsed_body=await _read_request_body(request=request),
|
||||
numeric_fields=IMAGE_EDIT_NUMERIC_FORM_FIELDS,
|
||||
).items()
|
||||
if key not in BRACKETED_FILE_FIELDS
|
||||
}
|
||||
parsed_body: Final = coerce_numeric_form_fields(
|
||||
parsed_body=await _read_request_body(request=request),
|
||||
numeric_fields=IMAGE_EDIT_NUMERIC_FORM_FIELDS,
|
||||
)
|
||||
for _field in BRACKETED_FILE_FIELDS:
|
||||
_value = parsed_body.get(_field)
|
||||
if isinstance(_value, str) or (isinstance(_value, list) and any(isinstance(_v, str) for _v in _value)):
|
||||
raise HTTPException(
|
||||
status_code=422,
|
||||
detail=f"'{_field}' must be provided as a multipart file upload, not a string.",
|
||||
)
|
||||
data: Final = {key: value for key, value in parsed_body.items() if key not in BRACKETED_FILE_FIELDS}
|
||||
image_files: Final = await batch_to_bytesio(image)
|
||||
mask_files: Final = await batch_to_bytesio(mask)
|
||||
if image_files:
|
||||
|
|
|
|||
|
|
@ -0,0 +1,151 @@
|
|||
"""Regression test for BerriAI/litellm#41421.
|
||||
|
||||
Proxy `/v1/images/edits` must forward `image[]`/`mask[]` multipart uploads as
|
||||
file bytes, never as `str()`-serialized objects. Previously `_read_request_body`
|
||||
left the collapsed `image[]` UploadFile in `data` alongside the converted
|
||||
`image` BytesIO list, and downstream `flatten_form_field_values` stringified
|
||||
the leftover into a form field OpenAI rejected with
|
||||
"Invalid type for 'image[0]': expected a file, but got a string instead."
|
||||
"""
|
||||
|
||||
import io
|
||||
from typing import Any
|
||||
|
||||
from fastapi import FastAPI
|
||||
from fastapi.testclient import TestClient
|
||||
|
||||
from litellm.proxy._types import UserAPIKeyAuth
|
||||
from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
|
||||
from litellm.proxy.image_endpoints import endpoints
|
||||
|
||||
PNG_BYTES = b"\x89PNG\r\n\x1a\n" + b"\x00" * 100
|
||||
|
||||
|
||||
def _client(monkeypatch, captured: dict[str, Any]) -> TestClient:
|
||||
class CaptureProcessing:
|
||||
def __init__(self, data: dict[str, Any]) -> None:
|
||||
captured.update(data)
|
||||
|
||||
async def base_process_llm_request(self, **_: Any) -> dict[str, Any]:
|
||||
return {"data": [{"b64_json": "aGk="}]}
|
||||
|
||||
monkeypatch.setattr(endpoints, "ProxyBaseLLMRequestProcessing", CaptureProcessing)
|
||||
monkeypatch.setattr("litellm.proxy.proxy_server.general_settings", {})
|
||||
monkeypatch.setattr("litellm.proxy.proxy_server.user_model", None)
|
||||
|
||||
app = FastAPI()
|
||||
app.include_router(endpoints.router)
|
||||
app.dependency_overrides[user_api_key_auth] = lambda: UserAPIKeyAuth()
|
||||
return TestClient(app)
|
||||
|
||||
|
||||
def test_image_bracket_upload_stays_bytes_not_str(monkeypatch):
|
||||
"""`image[]` file must reach the provider as BytesIO with no `image[]` leftover."""
|
||||
captured: dict[str, Any] = {}
|
||||
response = _client(monkeypatch, captured).post(
|
||||
"/v1/images/edits",
|
||||
files={"image[]": ("input.png", PNG_BYTES, "image/png")},
|
||||
data={"model": "openai/gpt-image-2", "prompt": "Make this image photorealistic"},
|
||||
)
|
||||
|
||||
assert response.status_code == 200
|
||||
# No raw multipart key may leak downstream where flatten would str() it.
|
||||
assert "image[]" not in captured
|
||||
assert "mask[]" not in captured
|
||||
images = captured["image"]
|
||||
assert isinstance(images, list) and len(images) == 1
|
||||
assert isinstance(images[0], io.BytesIO)
|
||||
assert not isinstance(images[0], str)
|
||||
assert not any(isinstance(v, str) and "BytesIO object at" in v for v in images)
|
||||
images[0].seek(0)
|
||||
assert images[0].read() == PNG_BYTES
|
||||
assert getattr(images[0], "name", None) == "input.png"
|
||||
|
||||
|
||||
def test_mask_bracket_upload_stays_bytes_not_str(monkeypatch):
|
||||
"""`mask[]` file must reach the provider as BytesIO with no `mask[]` leftover."""
|
||||
captured: dict[str, Any] = {}
|
||||
response = _client(monkeypatch, captured).post(
|
||||
"/v1/images/edits",
|
||||
files={
|
||||
"image": ("input.png", PNG_BYTES, "image/png"),
|
||||
"mask[]": ("mask.png", PNG_BYTES, "image/png"),
|
||||
},
|
||||
data={"model": "openai/gpt-image-2", "prompt": "hi"},
|
||||
)
|
||||
|
||||
assert response.status_code == 200
|
||||
assert "image[]" not in captured
|
||||
assert "mask[]" not in captured
|
||||
masks = captured["mask"]
|
||||
assert isinstance(masks, list) and len(masks) == 1
|
||||
assert isinstance(masks[0], io.BytesIO)
|
||||
masks[0].seek(0)
|
||||
assert masks[0].read() == PNG_BYTES
|
||||
|
||||
|
||||
def test_multiple_bracket_uploads_all_reach_provider(monkeypatch):
|
||||
"""Repeated `image[]` parts must all arrive (form-dict collapsing must not drop any)."""
|
||||
captured: dict[str, Any] = {}
|
||||
client = _client(monkeypatch, captured)
|
||||
png2 = b"\x89PNG\r\n\x1a\n" + b"\x01" * 100
|
||||
response = client.post(
|
||||
"/v1/images/edits",
|
||||
files=[
|
||||
("image[]", ("a.png", PNG_BYTES, "image/png")),
|
||||
("image[]", ("b.png", png2, "image/png")),
|
||||
],
|
||||
data={"model": "openai/gpt-image-2", "prompt": "hi"},
|
||||
)
|
||||
|
||||
assert response.status_code == 200
|
||||
assert "image[]" not in captured
|
||||
assert len(captured["image"]) == 2
|
||||
assert [getattr(b, "name", None) for b in captured["image"]] == ["a.png", "b.png"]
|
||||
|
||||
|
||||
def test_string_image_upload_rejected(monkeypatch):
|
||||
"""A string `image`/`image[]` field must 422, not forward as a string file."""
|
||||
for field in ("image", "image[]"):
|
||||
captured: dict[str, Any] = {}
|
||||
response = _client(monkeypatch, captured).post(
|
||||
"/v1/images/edits",
|
||||
data={"model": "openai/gpt-image-2", "prompt": "hi", field: "not-a-file"},
|
||||
)
|
||||
assert response.status_code == 422, field
|
||||
|
||||
|
||||
def test_numeric_coercion_preserved_with_file_upload(monkeypatch):
|
||||
"""Numeric coercion (`n` -> int) must keep working alongside the file fix."""
|
||||
captured: dict[str, Any] = {}
|
||||
response = _client(monkeypatch, captured).post(
|
||||
"/v1/images/edits",
|
||||
files={"image[]": ("input.png", PNG_BYTES, "image/png")},
|
||||
data={"model": "openai/gpt-image-2", "prompt": "hi", "n": "2"},
|
||||
)
|
||||
|
||||
assert response.status_code == 200
|
||||
assert captured["n"] == 2 and isinstance(captured["n"], int)
|
||||
assert isinstance(captured["image"][0], io.BytesIO)
|
||||
|
||||
|
||||
def test_transformation_preserves_bytesio_filename():
|
||||
"""BytesIO `.name` set by the proxy must survive into the httpx files tuple."""
|
||||
from litellm.llms.openai.image_edit.transformation import OpenAIImageEditConfig
|
||||
from litellm.types.router import GenericLiteLLMParams
|
||||
|
||||
buf = io.BytesIO(PNG_BYTES)
|
||||
buf.name = "input.png"
|
||||
data, files = OpenAIImageEditConfig().transform_image_edit_request(
|
||||
model="gpt-image-2",
|
||||
prompt="hi",
|
||||
image=[buf],
|
||||
image_edit_optional_request_params={},
|
||||
litellm_params=GenericLiteLLMParams(),
|
||||
headers={},
|
||||
)
|
||||
assert "image" not in data # files stay out of the data dict
|
||||
assert files[0][0] == "image[]"
|
||||
assert files[0][1][0] == "input.png"
|
||||
assert files[0][1][1] is buf
|
||||
assert isinstance(files[0][1][1], io.BytesIO)
|
||||
Loading…
Add table
Reference in a new issue