fn.py

import os
import io
import base64
import torch
from torch.cuda import amp
import numpy as np
from PIL import Image
from diffusers import AutoPipelineForText2Image, AutoencoderKL, DPMSolverMultistepScheduler
from diffusers import StableDiffusion3Pipeline

pipe = None

def load_model(_model = None, _vae = None, loras = []):
    global pipe

    _model = _model or "v2ray/stable-diffusion-3-medium-diffusers"

    if torch.cuda.is_available():
        torch_dtype = torch.float16
    else:
        torch_dtype = torch.float32

    kwargs = {}
    if _vae:
        # "stabilityai/sdxl-vae"
        vae = AutoencoderKL.from_pretrained(_vae, torch_dtype=torch_dtype)
        kwargs['vae'] = vae

    pipe = StableDiffusion3Pipeline.from_pretrained(_model, torch_dtype=torch_dtype, **kwargs)

    # DPM++ 2M Karras
    # pipe.scheduler = DPMSolverMultistepScheduler.from_config(
        # pipe.scheduler.config,
        # algorithm_type="sde-dpmsolver++",
        # use_karras_sigmas=True
        # )

    for lora in loras:
        pipe.load_lora_weights(".", weight_name=lora + ".safetensors")

    if torch.cuda.is_available():
        pipe = pipe.to("cuda")

    #pipe.enable_vae_slicing()

def pil_to_webp(img):
    buffer = io.BytesIO()
    img.save(buffer, 'webp')

    return buffer.getvalue()

def bin_to_base64(bin):
    return base64.b64encode(bin).decode('ascii')

def run(prompt = None, negative_prompt = None, model = None, guidance_scale = None, steps = None, seed = None):
    global pipe
    
    if not pipe:
        load_model(model)

    _prompt = "A cat holding a sign that says hello world"
    _negative_prompt = ""

    prompt = prompt or _prompt
    negative_prompt = negative_prompt or _negative_prompt
    guidance_scale = float(guidance_scale) if guidance_scale else 7.0
    steps = int(steps) if steps else 28
    seed = int(seed) if seed else -1

    generator = None
    if seed != -1:
        generator = torch.manual_seed(seed)

    image = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        guidance_scale=guidance_scale,
        num_inference_steps=steps,
        clip_skip=2,
        generator=generator,
        ).images[0]

    return image