|
from typing import Dict, List, Any |
|
import torch |
|
import requests |
|
from PIL import Image |
|
from io import BytesIO |
|
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline, DDIMScheduler |
|
|
|
|
|
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') |
|
|
|
if device.type != 'cuda': |
|
raise ValueError("need to run on GPU") |
|
|
|
model_id = "stabilityai/stable-diffusion-2-1-base" |
|
|
|
class EndpointHandler(): |
|
def __init__(self, path=""): |
|
|
|
self.textPipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16) |
|
self.textPipe.scheduler = DDIMScheduler.from_config(self.textPipe.scheduler.config) |
|
self.textPipe = self.textPipe.to(device) |
|
|
|
|
|
self.imgPipe = StableDiffusionImg2ImgPipeline.from_pretrained(model_id, torch_dtype=torch.float16) |
|
self.imgPipe.scheduler = DDIMScheduler.from_config(self.imgPipe.scheduler.config) |
|
self.imgPipe = self.imgPipe.to(device) |
|
|
|
def __call__(self, data: Any) -> List[List[Dict[str, float]]]: |
|
""" |
|
Args: |
|
data (:obj:): |
|
includes the input data and the parameters for the inference. |
|
Return: |
|
A :obj:`dict`:. base64 encoded image |
|
""" |
|
prompt = data.pop("inputs", data) |
|
url = data.pop("src", data) |
|
response = requests.get(url) |
|
init_image = Image.open(BytesIO(response.content)).convert("RGB") |
|
init_image.thumbnail((512, 512)) |
|
|
|
params = data.pop("parameters", data) |
|
|
|
|
|
num_inference_steps = params.pop("num_inference_steps", 25) |
|
guidance_scale = params.pop("guidance_scale", 7.5) |
|
negative_prompt = params.pop("negative_prompt", None) |
|
height = params.pop("height", None) |
|
width = params.pop("width", None) |
|
manual_seed = params.pop("manual_seed", -1) |
|
|
|
out = None |
|
|
|
if data.get("src"): |
|
generator = torch.Generator(device='cuda') |
|
generator.manual_seed(manual_seed) |
|
|
|
out = self.imgPipe(prompt, |
|
image=init_image, |
|
num_inference_steps=num_inference_steps, |
|
guidance_scale=guidance_scale, |
|
num_images_per_prompt=1, |
|
negative_prompt=negative_prompt, |
|
height=height, |
|
width=width |
|
) |
|
else: |
|
|
|
out = self.textPipe(prompt, |
|
image=init_image, |
|
num_inference_steps=num_inference_steps, |
|
guidance_scale=guidance_scale, |
|
num_images_per_prompt=1, |
|
negative_prompt=negative_prompt, |
|
height=height, |
|
width=width |
|
) |
|
|
|
|
|
|
|
return out.images[0] |
|
|
