app.py

from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
from diffusers import UniPCMultistepScheduler
import gradio as gr
import numpy as np
import torch
import base64
import cv2
from io import BytesIO
from PIL import Image, ImageFilter

# Constants
low_threshold = 100
high_threshold = 200

canvas_html = '<pose-maker/>'
load_js = """
async () => {
  const url = "https://huggingface.co/datasets/mishig/gradio-components/raw/main/mannequinAll.js"
  fetch(url)
    .then(res => res.text())
    .then(text => {
      const script = document.createElement('script');
      script.type = "module"
      script.src = URL.createObjectURL(new Blob([text], { type: 'application/javascript' }));
      document.head.appendChild(script);
    });
}
"""

get_js_image = """
async (canvas, prompt) => {
  const poseMakerEl = document.querySelector("pose-maker");
  const imgBase64 = poseMakerEl.captureScreenshot();
  return [imgBase64, prompt]
}
"""

# Models
controlnet = ControlNetModel.from_pretrained(
    "lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16
)
pipe = StableDiffusionControlNetPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None, torch_dtype=torch.float16
)
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)

# This command loads the individual model components on GPU on-demand. So, we don't
# need to explicitly call pipe.to("cuda").
pipe.enable_model_cpu_offload()

# xformers
pipe.enable_xformers_memory_efficient_attention()

# Generator seed,
generator = torch.manual_seed(0)

def get_canny_filter(image):
    if not isinstance(image, np.ndarray):
        image = np.array(image) 
        
    image = cv2.Canny(image, low_threshold, high_threshold)
    image = image[:, :, None]
    image = np.concatenate([image, image, image], axis=2)
    canny_image = Image.fromarray(image)
    return canny_image

def generate_images(canvas, prompt):
    try:
        base64_img = canvas
        image_data = base64.b64decode(base64_img.split(',')[1])
        input_img = Image.open(BytesIO(image_data)).convert(
            'RGB').resize((512, 512))
        input_img = input_img.filter(ImageFilter.GaussianBlur(radius=2))
        input_img = get_canny_filter(input_img)
        output = pipe(
            f'{prompt}, best quality, extremely detailed',
            input_img,
            generator=generator,
            num_images_per_prompt=2,
            num_inference_steps=20,
            negative_prompt="longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality",
        )
        all_outputs = [input_img]
        for image in output.images:
            all_outputs.append(image)
        return all_outputs
    except Exception as e:
        raise gr.Error(str(e))

def placeholder_fn(axis):
    pass

js_change_rotation_axis = """
async (axis) => {
  const poseMakerEl = document.querySelector("pose-maker");
  poseMakerEl.changeRotationAxis(axis);
}
"""

js_pose_template = """
async (pose) => {
  const poseMakerEl = document.querySelector("pose-maker");
  poseMakerEl.setPose(pose);
}
"""

with gr.Blocks() as blocks:
    gr.HTML(
        """
            <div style="text-align: center; margin: 0 auto;">
              <div
                style="
                  display: inline-flex;
                  align-items: center;
                  gap: 0.8rem;
                  font-size: 1.75rem;
                "
              >
                <h1 style="font-weight: 900; margin-bottom: 7px;margin-top:5px">
                  Pose in 3D & Render with ControlNet (SD-1.5)
                </h1>
              </div>
              <p style="margin-bottom: 10px; font-size: 94%; line-height: 23px;">
                Using <a href="https://github.com/lllyasviel/ControlNet">ControlNet</a> and <a href="https://boytchev.github.io/mannequin.js/">three.js/mannequin.js</a>
              </p>
              <p>For faster inference without waiting in queue, you may duplicate the space and upgrade to GPU in settings. <a href="https://huggingface.co/spaces/diffusers/controlnet-3d-pose?duplicate=true"><img style="display: inline; margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space" /></a></p>
            </div>
        """
    )
    with gr.Row():
        with gr.Column():
            canvas = gr.HTML(canvas_html, elem_id="canvas_html", visible=True)
            with gr.Row():
                rotation_axis = gr.Radio(["x", "y", "z"], value="x", label="Joint rotation axis")
                pose_template = gr.Radio(["regular", "ballet", "handstand", "split", "kick", "chilling"], value="regular", label="Pose template")
            prompt = gr.Textbox(
                label="Enter your prompt",
                max_lines=1,
                placeholder="best quality, extremely detailed",
            )
            run_button = gr.Button("Generate")
        with gr.Column():
            gallery = gr.Gallery().style(grid=[2], height="auto")
    rotation_axis.change(fn=placeholder_fn,
                            inputs=[rotation_axis],
                            outputs=[],
                            queue=False,
                            _js=js_change_rotation_axis)
    pose_template.change(fn=placeholder_fn,
                            inputs=[pose_template],
                            outputs=[],
                            queue=False,
                            _js=js_pose_template)
    run_button.click(fn=generate_images,
                     inputs=[canvas, prompt],
                     outputs=[gallery],
                     _js=get_js_image)
    blocks.load(None, None, None, _js=load_js)

blocks.launch(debug=True)