model.py

import torch
from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler, DPMSolverMultistepScheduler, \
    OnnxStableDiffusionPipeline

import pipeline_openvino_stable_diffusion
from optimum.intel.openvino import OVStableDiffusionPipeline


def get_sd_21():
    model_id = "stabilityai/stable-diffusion-2-1-base"
    scheduler = EulerDiscreteScheduler.from_pretrained(model_id, subfolder="scheduler")

    if torch.cuda.is_available():
        pipe = StableDiffusionPipeline.from_pretrained(
            model_id,
            scheduler=scheduler,
            # safety_checker=None,
            revision="fp16",
            torch_dtype=torch.float16)
        pipe = pipe.to('cuda')
    else:
        pipe = StableDiffusionPipeline.from_pretrained(
            model_id,
            scheduler=scheduler,
            # safety_checker=None,
            revision="fp16",
            torch_dtype=torch.float16)
    return pipe


def get_sd_every():
    model_id = 'OFA-Sys/small-stable-diffusion-v0'
    scheduler = DPMSolverMultistepScheduler.from_pretrained(model_id, subfolder="scheduler")
    pipe = OVStableDiffusionPipeline.from_pretrained("OpenVINO/stable-diffusion-2-1-quantized", compile=False)
    pipe.compile()
    return pipe


def get_sd_small():
    pipe = OVStableDiffusionPipeline.from_pretrained("OpenVINO/stable-diffusion-2-1-quantized", compile=False)
    pipe.reshape(batch_size=1, height=256, width=256, num_images_per_prompt=1)
    pipe.compile()
    return pipe


def get_sd_tiny():
    pipe = OVStableDiffusionPipeline.from_pretrained("OpenVINO/stable-diffusion-2-1-quantized", compile=False)
    pipe.reshape(batch_size=1, height=512, width=512, num_images_per_prompt=1)
    pipe.compile()
    return pipe