app.py

import cv2, os, math
import torch
import random
import numpy as np

import spaces

import PIL
from PIL import Image
from typing import Tuple

import diffusers
from diffusers.utils import load_image

from diffusers import (
    AutoencoderKL,
    UNet2DConditionModel,
    UniPCMultistepScheduler,
)

from huggingface_hub import hf_hub_download

from insightface.app import FaceAnalysis

from pipeline_controlnet_xs_sd_xl_instantid import StableDiffusionXLInstantIDXSPipeline, UNetControlNetXSModel

from utils.controlnet_xs import ControlNetXSAdapter
from style import styles


import gradio as gr

hf_hub_download(repo_id="RED-AIGC/InstantID-XS", filename="controlnetxs.bin", local_dir="./ckpt")
hf_hub_download(repo_id="RED-AIGC/InstantID-XS",filename="cross_attn.bin",local_dir="./ckpt",)
hf_hub_download(repo_id="RED-AIGC/InstantID-XS", filename="image_proj.bin", local_dir="./ckpt")


# global variable
MAX_SEED = np.iinfo(np.int32).max
device = "cuda" if torch.cuda.is_available() else "cpu"
weight_dtype = torch.float16 if str(device).__contains__("cuda") else torch.float32
STYLE_NAMES = list(styles.keys())
DEFAULT_STYLE_NAME = "Ordinary"

base_model = 'frankjoshua/realvisxlV40_v40Bakedvae'
vae_path = 'madebyollin/sdxl-vae-fp16-fix'
# ckpt = 'RED-AIGC/InstantID-XS'

image_proj_path = "./ckpt/image_proj.bin"
cnxs_path =  "./ckpt/controlnetxs.bin"
cross_attn_path = "./ckpt/cross_attn.bin"


# Load face encoder
app = FaceAnalysis(
    name="antelopev2",
    root="./",
    providers=["CPUExecutionProvider"],
)
app.prepare(ctx_id=0, det_size=(640, 640))


def get_ControlNetXS(base_model, cnxs_path, device, size_ratio=0.125, weight_dtype=torch.float16):
    unet = UNet2DConditionModel.from_pretrained(base_model, subfolder="unet").to(device, dtype=weight_dtype)
    controlnet = ControlNetXSAdapter.from_unet(unet, size_ratio=size_ratio, learn_time_embedding=True)
    state_dict = torch.load(cnxs_path, map_location="cpu", weights_only=True)
    ctrl_state_dict = {}
    for key, value in state_dict.items():
        if 'attn2.processor' not in key:
            if 'ctrl_' in key and 'ctrl_to_base' not in key:
                key = key.replace('ctrl_', '')
            if 'up_blocks' in key:
                key = key.replace('up_blocks', 'up_connections')
            ctrl_state_dict[key] = value
    controlnet.load_state_dict(ctrl_state_dict, strict=True)
    controlnet.to(device, dtype=weight_dtype)
    ControlNetXS = UNetControlNetXSModel.from_unet(unet, controlnet).to(device, dtype=weight_dtype)
    
    return ControlNetXS

print('Get ControlNetXS...')
ControlNetXS = get_ControlNetXS(base_model, cnxs_path, device, size_ratio=0.125, weight_dtype=weight_dtype)

vae = AutoencoderKL.from_pretrained(vae_path)

print('Get Pipeline...')
pipe = StableDiffusionXLInstantIDXSPipeline.from_pretrained(
    base_model,
    vae=vae,
    unet=ControlNetXS,
    controlnet=None,
    torch_dtype=weight_dtype,
)

# pipe.cuda(device=device, dtype=weight_dtype, use_xformers=True)
pipe.cuda(device=device, dtype=weight_dtype, use_xformers=False)

print('Load IP-Adapter...')
pipe.load_ip_adapter(image_proj_path, cross_attn_path)

pipe.scheduler = diffusers.EulerDiscreteScheduler.from_config(pipe.scheduler.config)
pipe.unet.config.ctrl_learn_time_embedding = True
pipe = pipe.to(device)


def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    return seed

def remove_tips():
    return gr.update(visible=False)

def get_example():
    case = [
        [
            "./examples/1.jpg",
            None,
            "Ordinary",
            "a woman",
            "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
        ],
        [
            "./examples/1.jpg",
            "./examples/pose/pose1.jpg",
            "Hanfu",
            "a woman",
            "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
        ],
        [
            "./examples/2.jpg",
            "./examples/pose/pose2.png",
            "ZangZu",
            "a woman",
            "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
        ],
        [
            "./examples/3.png",
            "./examples/pose/pose3.png",
            "QingQiu",
            "a woman",
            "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
        ],
    ]
    return case


def convert_from_cv2_to_image(img: np.ndarray) -> Image:
    return Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))

def convert_from_image_to_cv2(img: Image) -> np.ndarray:
    return cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)


def draw_kps(image_pil, kps, color_list=[(255,0,0), (0,255,0), (0,0,255), (255,255,0), (255,0,255)]):
    
    stickwidth = 4
    limbSeq = np.array([[0, 2], [1, 2], [3, 2], [4, 2]])
    kps = np.array(kps)

    w, h = image_pil.size
    out_img = np.zeros([h, w, 3])

    for i in range(len(limbSeq)):
        index = limbSeq[i]
        color = color_list[index[0]]

        x = kps[index][:, 0]
        y = kps[index][:, 1]
        length = ((x[0] - x[1]) ** 2 + (y[0] - y[1]) ** 2) ** 0.5
        angle = math.degrees(math.atan2(y[0] - y[1], x[0] - x[1]))
        polygon = cv2.ellipse2Poly((int(np.mean(x)), int(np.mean(y))), (int(length / 2), stickwidth), int(angle), 0, 360, 1)
        out_img = cv2.fillConvexPoly(out_img.copy(), polygon, color)
    out_img = (out_img * 0.6).astype(np.uint8)

    for idx_kp, kp in enumerate(kps):
        color = color_list[idx_kp]
        x, y = kp
        out_img = cv2.circle(out_img.copy(), (int(x), int(y)), 10, color, -1)

    out_img_pil = PIL.Image.fromarray(out_img.astype(np.uint8))
    return out_img_pil

def resize_img(input_image,max_side=1280,min_side=1024,size=None,pad_to_max_side=False,mode=PIL.Image.BILINEAR,base_pixel_number=64,):
    w, h = input_image.size
    if size is not None:
        w_resize_new, h_resize_new = size
    else:
        ratio = min_side / min(h, w)
        w, h = round(ratio * w), round(ratio * h)
        ratio = max_side / max(h, w)
        input_image = input_image.resize([round(ratio * w), round(ratio * h)], mode)
        w_resize_new = (round(ratio * w) // base_pixel_number) * base_pixel_number
        h_resize_new = (round(ratio * h) // base_pixel_number) * base_pixel_number
    input_image = input_image.resize([w_resize_new, h_resize_new], mode)

    if pad_to_max_side:
        res = np.ones([max_side, max_side, 3], dtype=np.uint8) * 255
        offset_x = (max_side - w_resize_new) // 2
        offset_y = (max_side - h_resize_new) // 2
        res[
            offset_y : offset_y + h_resize_new, offset_x : offset_x + w_resize_new
        ] = np.array(input_image)
        input_image = Image.fromarray(res)
    return input_image

def apply_style(style_name: str, positive: str, negative: str = "") -> Tuple[str, str]:
    p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
    return p.replace("{prompt}", positive), n + ' ' + negative

def run_for_examples(face_file, pose_file, style, prompt, negative_prompt, ):
    return generate_image(
        face_file,
        pose_file,
        style, 
        prompt,
        negative_prompt,
        20,  # num_steps
        0.9,  # ControlNet strength
        0.8,  # Adapter strength
        5.0,  # guidance_scale
        42,  # seed
        1280, # max side
    )

@spaces.GPU
def generate_image(
    face_image_path,
    pose_image_path,
    style_name, 
    prompt,
    negative_prompt,
    num_steps,
    controlnet_conditioning_scale,
    adapter_strength_ratio,
    guidance_scale,
    seed,
    max_side,
    progress=gr.Progress(track_tqdm=True),
):

    if face_image_path is None:
        raise gr.Error(f"Cannot find any input face image! Please upload the face image")

    if prompt is None:
        prompt = "a person"

    # apply the style template
    prompt, negative_prompt = apply_style(style_name, prompt, negative_prompt)

    face_image = load_image(face_image_path)
    face_image = resize_img(face_image, max_side=max_side)
    # face_image = resize_img(face_image)
    face_image_cv2 = convert_from_image_to_cv2(face_image)
    height, width, _ = face_image_cv2.shape

    # Extract face features
    face_info = app.get(face_image_cv2)

    if len(face_info) == 0:
        raise gr.Error(f"Unable to detect a face in the image. Please upload a different photo with a clear face.")

    face_info = sorted(
        face_info,
        key=lambda x: (x["bbox"][2] - x["bbox"][0]) * x["bbox"][3] - x["bbox"][1],
    )[-1]  # only use the maximum face

    face_emb = torch.from_numpy(face_info.normed_embedding)
    face_kps = draw_kps(convert_from_cv2_to_image(face_image_cv2), face_info["kps"])
    if pose_image_path is not None:
        pose_image = load_image(pose_image_path)
        pose_image = resize_img(pose_image, max_side=max_side)
        # pose_image = resize_img(pose_image)
        pose_image_cv2 = convert_from_image_to_cv2(pose_image)

        face_info = app.get(pose_image_cv2)

        if len(face_info) == 0:
            raise gr.Error(f"Cannot find any face in the reference image! Please upload another person image")

        face_info = face_info[-1]
        face_kps = draw_kps(pose_image, face_info["kps"])

        width, height = face_kps.size
    print(width, height)
    print("Start inference...")
    print(f"[Debug] Prompt: {prompt}, \n[Debug] Neg Prompt: {negative_prompt}")

    # pipe.set_ip_adapter_scale(adapter_strength_ratio)
    images = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        image=face_kps,
        face_emb=face_emb,
        controlnet_conditioning_scale=float(controlnet_conditioning_scale),
        ip_adapter_scale=float(adapter_strength_ratio),
        num_inference_steps=num_steps,
        guidance_scale=float(guidance_scale),
        height=height,
        width=width,
        generator=torch.Generator(device=device).manual_seed(seed),
    ).images

    return images[0], gr.update(visible=True)

title = r"""
<h1 align="center">InstantID-XS</h1>
"""

tips = r"""
### Usage tips of InstantID-XS
1. If you're not satisfied with the similarity, try increasing the weight of "ControlNet strength" and "Adapter Strength."    
2. If you feel that the similarity is not high, you can increase the adapter strength appropriately. 
3. If you want to achieve a pose image as similar as possible, please increase the ControlNet strength appropriately.
"""
css = """
.gradio-container {width: 85% !important}
"""

with gr.Blocks(css=css) as demo:
    # description
    gr.Markdown(title)
    # gr.Markdown(description)

    with gr.Row():
        with gr.Column():
            with gr.Row(equal_height=True):
                # upload face image
                face_file = gr.Image(label="Upload a photo of your face", type="filepath")
                # optional: upload a reference pose image
                pose_file = gr.Image(label="Upload a reference pose image (Optional)",type="filepath",)

            # prompt
            prompt = gr.Textbox(
                label="Prompt",
                info="Give simple prompt is enough to achieve good face fidelity",
                placeholder="A photo of a person",
                value="realistic, symmetrical hyperdetailed texture, masterpiece, enhanced details, perfect composition, authentic, natural posture",
            )

            submit = gr.Button("Submit", variant="primary")

            style = gr.Dropdown(
                label="Style", 
                choices=STYLE_NAMES, 
                value=DEFAULT_STYLE_NAME
            )
                
            # strength
            controlnet_conditioning_scale = gr.Slider(
                label="ControlNet strength (for pose)",
                minimum=0.0,
                maximum=1.0,
                step=0.1,
                value=0.9,
            )
            adapter_strength_ratio = gr.Slider(
                label="Adapter strength (for fidelity)",
                minimum=0.0,
                maximum=1.0,
                step=0.1,
                value=0.8,
            )

            with gr.Accordion(open=True, label="Advanced Options"):
                negative_prompt = gr.Textbox(
                    label="Negative Prompt",
                    placeholder="low quality",
                    value="(lowres, low quality, worst quality:1.2), (text:1.2), nude, nsfw, watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
                )
                num_steps = gr.Slider(
                    label="Number of sample steps",
                    minimum=1,
                    maximum=100,
                    step=1,
                    value=20,
                ) 
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=0.1,
                    maximum=10.0,
                    step=0.1,
                    value=5.0,
                )
                seed = gr.Slider(
                    label="Seed",
                    minimum=0,
                    maximum=MAX_SEED,
                    step=1,
                    value=42,
                )
                max_side = gr.Slider(
                    label="Max side",
                    minimum=512,
                    maximum=2048,
                    step=64,
                    value=1280,
                )

                randomize_seed = gr.Checkbox(label="Randomize seed", value=True)

        with gr.Column(scale=1):
            gallery = gr.Image(label="Generated Images")
            usage_tips = gr.Markdown(label="InstantID Usage Tips", value=tips, visible=False)

        submit.click(
            fn=remove_tips,
            outputs=usage_tips,
        ).then(
            fn=randomize_seed_fn,
            inputs=[seed, randomize_seed],
            outputs=seed,
            queue=False,
            api_name=False,
        ).then(
            fn=generate_image,
            inputs=[
                face_file,
                pose_file,
                style,
                prompt,
                negative_prompt,
                num_steps,
                controlnet_conditioning_scale,
                adapter_strength_ratio,
                guidance_scale,
                seed,
                max_side,
            ],
            outputs=[gallery, usage_tips],
        )

    gr.Examples(
        examples=get_example(),
        inputs=[face_file, pose_file, style, prompt, negative_prompt],
        fn=run_for_examples,
        outputs=[gallery, usage_tips],
        cache_examples=True,
    )

    # gr.Markdown(article)

demo.queue(api_open=False)
demo.launch()