modules/default_pipeline.py

import os
import torch
import modules.core as core

from modules.path import modelfile_path


xl_base_filename = os.path.join(modelfile_path, 'sd_xl_base_1.0.safetensors')
xl_refiner_filename = os.path.join(modelfile_path, 'sd_xl_refiner_1.0.safetensors')

xl_base = core.load_model(xl_base_filename)
xl_refiner = core.load_model(xl_refiner_filename)
del xl_base.vae


@torch.no_grad()
def process(positive_prompt, negative_prompt, width=1024, height=1024, batch_size=1):
    positive_conditions = core.encode_prompt_condition(clip=xl_base.clip, prompt=positive_prompt)
    negative_conditions = core.encode_prompt_condition(clip=xl_base.clip, prompt=negative_prompt)

    positive_conditions_refiner = core.encode_prompt_condition(clip=xl_refiner.clip, prompt=positive_prompt)
    negative_conditions_refiner = core.encode_prompt_condition(clip=xl_refiner.clip, prompt=negative_prompt)

    empty_latent = core.generate_empty_latent(width=width, height=height, batch_size=batch_size)

    sampled_latent = core.ksampler_with_refiner(
        model=xl_base.unet,
        positive=positive_conditions,
        negative=negative_conditions,
        refiner=xl_refiner.unet,
        refiner_positive=positive_conditions_refiner,
        refiner_negative=negative_conditions_refiner,
        refiner_switch_step=20,
        latent=empty_latent,
        steps=30, start_step=0, last_step=30, disable_noise=False, force_full_denoise=True,
        seed=12345
    )

    decoded_latent = core.decode_vae(vae=xl_refiner.vae, latent_image=sampled_latent)

    images = core.image_to_numpy(decoded_latent)

    core.close_all_preview()

    return images