from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler, AutoencoderKL
import torch
import os


#////////////////////////////////////////////////////////////////
guidance_scale=7
steps=20
width=1024
height=1024
base_model_id_str = "SDXL-xmasize-Lora-Images"
prompt_prefix = ""
prompt_suffix = " xmasize, Very detailed, clean, high quality, sharp image"
neg_prompt = "text, watermark, grainy, blurry, unfocused, nsfw, naked, nude, noisy image, deformed, distorted, pixelated"
#////////////////////////////////////////////////////////////////
base = None
refiner = None
#////////////////////////////////////////////////////////////////


def load():
    global base, refiner

    vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
    base = DiffusionPipeline.from_pretrained(
        "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
    )
    base.to("cuda")
    base.load_lora_weights("Norod78/SDXL-xmasize-Lora")
    base.enable_xformers_memory_efficient_attention()
    refiner = DiffusionPipeline.from_pretrained(
        "stabilityai/stable-diffusion-xl-refiner-1.0",
        text_encoder_2=base.text_encoder_2,
        vae=base.vae,
        torch_dtype=torch.float16,
        use_safetensors=True,
        variant="fp16",
    )
    refiner.to("cuda")
    refiner.enable_xformers_memory_efficient_attention()

def generate(prompt, file_prefix ,samples = 2, seed = 7777):
    global base, refiner

    torch.manual_seed(seed)
    prompt = prompt_prefix + prompt
    prompt += prompt_suffix
    base_model_latents = base([prompt] * samples, 
                              negative_prompt = [neg_prompt] * samples, 
                              num_inference_steps=steps, 
                              guidance_scale=guidance_scale,
                              height=height, width=width,
                              output_type="latent")["images"]
    torch.manual_seed(seed)
    refiner_model_images = refiner([prompt] * samples, 
                                   negative_prompt = [neg_prompt] * samples, 
                                   num_inference_steps=steps, 
                                   image=base_model_latents)["images"]
    for idx, image in enumerate(refiner_model_images):
        image.save(f"{base_model_id_str}/{file_prefix}-{idx}-{seed}--{width}x{height}--{guidance_scale}--{base_model_id_str}.jpg")

def main():
    load()

    os.mkdir(base_model_id_str)

    generate("A livingroom", "01_LivingRoom")
    generate("A nice town", "02_NiceTown")
    generate("A scene in \"The Minions\" movie", "03_MinionsMovie")
    generate("Wonderwoman", "04_Wonderwoman")
    generate("Marge Simpson", "05_MargeSimpson")
    generate("A beautiful woman", "06_BeautifulWoman")
    generate("A magical landscape", "07_MagicalLandscape")
    generate("Cute dog", "08_CuteDog")
    generate("An oil on canvas portrait of Snoop Dogg, Mark Ryden", "09_SnoopDog")
    generate("A flemish baroque painting of Kermit from the muppet show", "10_KermitFlemishBaroque")
    generate("Gal Gadot in Avatar", "11_GalGadotAvatar")
    generate("Ninja turtles, Naoto Hattori", "12_TMNT")
    generate("A socially awkward potato", "13_AwkwardPotato")
    generate("Pikachu as Rick and morty, Eric Wallis", "14_PikachuRnM")
    generate("The girl with pearl earing", "15_PearlEaring")
    generate("American Gothic ", "16_AmericanGothic")
    generate("Miss. Piggy as the Mona Lisa", "17_MsPiggyMonaLisa")
    generate("Rick Sanchez from the TV show \"Rick and Morty\"", "18_RickSanchez")
    generate("A paiting of Southpark with rainbow", "19_Southpark")
    generate("An oil painting of Phineas and Pherb hamering on a new machine, Eric Wallis", "20_PhineasPherb")
    generate("Bender, Saturno Butto", "21_Bender")
    generate("A psychedelic image of Bojack Horseman", "22_Bojack")
    generate("A movie poster for Gravity Falls Cthulhu stories", "23_GravityFalls")
    generate("A vibrant oil painting portrait of She-Ra", "24_Shira", 2, 512)

if __name__ == '__main__':
    main()