mirror of
https://github.com/xtekky/gpt4free.git
synced 2025-10-05 08:16:58 +08:00
52 lines
1.4 KiB
Python
52 lines
1.4 KiB
Python
from __future__ import annotations
|
|
|
|
from typing import Optional
|
|
|
|
from .helper import format_image_prompt
|
|
from ..typing import AsyncResult, Messages
|
|
from .PollinationsAI import PollinationsAI
|
|
|
|
class PollinationsImage(PollinationsAI):
|
|
label = "Pollinations AI (Image)"
|
|
default_model = "flux"
|
|
default_vision_model = None
|
|
default_image_model = default_model
|
|
|
|
@classmethod
|
|
def get_models(cls, **kwargs):
|
|
if not cls.image_models:
|
|
cls.image_models = list(dict.fromkeys([*cls.image_models, *cls.extra_image_models]))
|
|
return cls.image_models
|
|
|
|
@classmethod
|
|
async def create_async_generator(
|
|
cls,
|
|
model: str,
|
|
messages: Messages,
|
|
proxy: str = None,
|
|
prompt: str = None,
|
|
width: int = 1024,
|
|
height: int = 1024,
|
|
seed: Optional[int] = None,
|
|
cache: bool = False,
|
|
nologo: bool = True,
|
|
private: bool = False,
|
|
enhance: bool = False,
|
|
safe: bool = False,
|
|
**kwargs
|
|
) -> AsyncResult:
|
|
async for chunk in cls._generate_image(
|
|
model=model,
|
|
prompt=format_image_prompt(messages, prompt),
|
|
proxy=proxy,
|
|
width=width,
|
|
height=height,
|
|
seed=seed,
|
|
cache=cache,
|
|
nologo=nologo,
|
|
private=private,
|
|
enhance=enhance,
|
|
safe=safe
|
|
):
|
|
yield chunk
|