import torch
import numpy as np
from PIL import Image
from diffusers import ControlNetModel, StableDiffusionXLControlNetImg2ImgPipeline, DDIMScheduler
from hidiffusion import apply_hidiffusion, remove_hidiffusion
import cv2 

controlnet = ControlNetModel.from_pretrained(
    "diffusers/controlnet-canny-sdxl-1.0", torch_dtype=torch.float16, variant="fp16"
).to("cuda")
scheduler = DDIMScheduler.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", subfolder="scheduler")

pipe = StableDiffusionXLControlNetImg2ImgPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    controlnet=controlnet,
    scheduler = scheduler,
    torch_dtype=torch.float16,
).to("cuda")

# Apply hidiffusion with a single line of code.
apply_hidiffusion(pipe)

pipe.enable_model_cpu_offload()
pipe.enable_xformers_memory_efficient_attention()

path = './assets/lara.jpeg'
ori_image = Image.open(path)
# get canny image
image = np.array(ori_image)
image = cv2.Canny(image, 50, 120)
image = image[:, :, None]
image = np.concatenate([image, image, image], axis=2)
canny_image = Image.fromarray(image)

controlnet_conditioning_scale = 0.5  # recommended for good generalization
prompt = "Lara Croft with brown hair, and is wearing a tank top, a brown backpack. The room is dark and has an old-fashioned decor with a patterned floor and a wall featuring a design with arches and a dark area on the right side, muted color, high detail, 8k high definition award winning"
negative_prompt = "underexposed, poorly drawn hands, duplicate hands, overexposed, bad art, beginner, amateur, abstract, disfigured, deformed, close up, weird colors, watermark"

image = pipe(prompt,
    image=ori_image,
    control_image=canny_image,
    height=1536,
    width=2048,
    strength=0.99,
    num_inference_steps=50,
    controlnet_conditioning_scale=controlnet_conditioning_scale,
    guidance_scale=12.5,
    negative_prompt = negative_prompt,
    eta=1.0
).images[0]

image.save("lara.jpg")