68 lines
3.8 KiB
Python
68 lines
3.8 KiB
Python
from typing import Optional, Union, List, Dict, Any
|
|
from pydantic import BaseModel, Field
|
|
from ..utils import BaseRequest, normalization_loras
|
|
|
|
|
|
|
|
class LoraWeightItem(BaseModel):
|
|
path: str = Field(..., description="Path to the LoRA model")
|
|
scale: float = Field(..., description="Scale of the LoRA model (range: 0.0 ~ 4.0)", ge=0.0, le=4.0)
|
|
|
|
# link api json(sdxl-lora.json)
|
|
class SdxlLora(BaseRequest):
|
|
"""
|
|
SDXL consists of an ensemble of experts pipeline for latent diffusion: In a first step,
|
|
the base model is used to generate (noisy) latents, which are then further processed
|
|
with a refinement model
|
|
"""
|
|
prompt: str = Field(..., description="Input prompt for image generation")
|
|
image: Optional[str] = Field(default="", description="The image for image-to-image generation.") # Description added based on common usage
|
|
mask_image: Optional[str] = Field(
|
|
default=None, description="The mask image tells the model where to generate new pixels (white) and where to preserve the original image (black). It acts as a stencil or guide for targeted image editing.")
|
|
strength: Optional[float] = Field(default=0.8, ge=0.01, le=1.0, step=0.01, description="Strength indicates extent to transform the reference image")
|
|
loras: Optional[List[LoraWeightItem]] = Field(None, description="List of LoRAs to apply (max 3)", max_items=3)
|
|
width: Optional[int] = Field(default=1024, description="Output image width", ge=512, le=1536)
|
|
height: Optional[int] = Field(default=1024, description="Output image height", ge=512, le=1536)
|
|
num_inference_steps: Optional[int] = Field(default=30, ge=1, le=50, description="Number of inference steps")
|
|
guidance_scale: Optional[float] = Field(default=5.0, ge=0.0, le=10.0, step=0.1, description="Guidance scale for generation")
|
|
num_images: Optional[int] = Field(default=1, ge=1, le=4, description="Number of images to generate")
|
|
seed: Optional[int] = Field(default=-1, description="Random seed (-1 for random)")
|
|
enable_safety_checker: Optional[bool] = Field(default=True, description="Enable safety checker")
|
|
|
|
def build_payload(self) -> dict:
|
|
"""Builds the request payload dictionary."""
|
|
payload = {
|
|
"prompt": self.prompt,
|
|
"image": self.image,
|
|
"mask_image": self.mask_image,
|
|
"strength": self.strength,
|
|
"loras": [lora.model_dump() for lora in self.loras] if self.loras else [],
|
|
"size": f"{self.width}*{self.height}",
|
|
"num_inference_steps": self.num_inference_steps,
|
|
"guidance_scale": self.guidance_scale,
|
|
"num_images": self.num_images,
|
|
"seed": self.seed,
|
|
"enable_safety_checker": self.enable_safety_checker,
|
|
}
|
|
if "loras" in payload and hasattr(self, "loras") and self.loras:
|
|
|
|
LORA_SCALE_MAX_FROM_JSON = 4.0
|
|
LORA_SCALE_DEFAULT_FROM_JSON = 1.0
|
|
|
|
|
|
loras_for_normalization = [lora.model_dump() for lora in self.loras]
|
|
payload["loras"] = normalization_loras(loras_for_normalization, LORA_SCALE_MAX_FROM_JSON, LORA_SCALE_DEFAULT_FROM_JSON)
|
|
return self._remove_empty_fields(payload)
|
|
|
|
def get_api_path(self):
|
|
"""Gets the API path for the request. Corresponds to api_path in the interface configuration json"""
|
|
return "/api/v3/wavespeed-ai/sdxl-lora"
|
|
|
|
def field_required(self):
|
|
"""Corresponds to required in the interface configuration json"""
|
|
return ["prompt"]
|
|
|
|
def field_order(self):
|
|
"""Corresponds to x-order-properties in the interface configuration json"""
|
|
return ["prompt", "image", "mask_image", "strength", "loras", "size", "num_inference_steps", "guidance_scale", "num_images", "seed", "enable_base64_output", "enable_safety_checker"]
|