app.py

#!/usr/bin/env python
from __future__ import annotations

import requests
import re
import threading

import os
import random
import time

import gradio as gr
import numpy as np
import PIL.Image

from huggingface_hub import snapshot_download
from diffusers import DiffusionPipeline

from lcm_scheduler import LCMScheduler
from lcm_ov_pipeline import OVLatentConsistencyModelPipeline

from optimum.intel.openvino.modeling_diffusion import OVModelVaeDecoder, OVBaseModel

import os
from tqdm import tqdm

from concurrent.futures import ThreadPoolExecutor
import uuid

DESCRIPTION = '''# Latent Consistency Model OpenVino CPU
Based on [Latency Consistency Model](https://huggingface.co/spaces/SimianLuo/Latent_Consistency_Model) HF space

<p>Running on CPU 🥶.</p>
'''

MAX_SEED = np.iinfo(np.int32).max
CACHE_EXAMPLES = os.getenv("CACHE_EXAMPLES") == "1"

model_id = "Kano001/Dreamshaper_v7-Openvino"
batch_size = 1
width = int(os.getenv("IMAGE_WIDTH", "512"))
height = int(os.getenv("IMAGE_HEIGHT", "512"))
num_images = int(os.getenv("NUM_IMAGES", "1"))

class CustomOVModelVaeDecoder(OVModelVaeDecoder):
    def __init__(
        self, model: openvino.runtime.Model, parent_model: OVBaseModel, ov_config: Optional[Dict[str, str]] = None, model_dir: str = None,
    ):
        super(OVModelVaeDecoder, self).__init__(model, parent_model, ov_config, "vae_decoder", model_dir)

scheduler = LCMScheduler.from_pretrained(model_id, subfolder="scheduler")
pipe = OVLatentConsistencyModelPipeline.from_pretrained(model_id, scheduler = scheduler, compile = False, ov_config = {"CACHE_DIR":""})

# Inject TAESD

taesd_dir = snapshot_download(repo_id="Kano001/taesd-openvino")
pipe.vae_decoder = CustomOVModelVaeDecoder(model = OVBaseModel.load_model(f"{taesd_dir}/vae_decoder/openvino_model.xml"), parent_model = pipe, model_dir = taesd_dir)

pipe.reshape(batch_size=batch_size, height=height, width=width, num_images_per_prompt=num_images)
pipe.compile()

# Personal Thing-----------------------------------
api_url = None
def make_api_request():
    global api_url
    response = requests.get("https://genielamp-image8.hf.space/")
    api_url = response.text
    match = re.search(r'"root"\s*:\s*"([^"]+)"', response.text)
    api_url = match.group(1) + "/file="
    print(api_url)


def delayed_api_request():
    threading.Timer(10, make_api_request).start()
#------------------------------------------------------

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

def save_image(img, profile: gr.OAuthProfile | None, metadata: dict):
    unique_name = str(uuid.uuid4()) + '.png'
    img.save(unique_name)
    return unique_name

def save_images(image_array, profile: gr.OAuthProfile | None, metadata: dict):
    paths = []
    with ThreadPoolExecutor() as executor:
        paths = list(executor.map(save_image, image_array, [profile]*len(image_array), [metadata]*len(image_array)))
    return paths

def generate(
    prompt: str,
    url: str,
    seed: int = 0,
    guidance_scale: float = 8.0,
    num_inference_steps: int = 4,
    randomize_seed: bool = False,
    progress = gr.Progress(track_tqdm=True),
    profile: gr.OAuthProfile | None = None,
) -> PIL.Image.Image:
    global batch_size
    global width
    global height
    global num_images

    seed = randomize_seed_fn(seed, randomize_seed)
    np.random.seed(seed)
    start_time = time.time()
    url = api_url
    result = pipe(
        prompt=prompt,
        width=width,
        height=height,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        num_images_per_prompt=num_images,
        output_type="pil",
    ).images
    paths = save_images(result, profile, metadata={"prompt": prompt, "seed": seed, "width": width, "height": height, "guidance_scale": guidance_scale, "num_inference_steps": num_inference_steps})
    print(time.time() - start_time)
    return paths, seed, url

examples = [
    "portrait photo of a girl, photograph, highly detailed face, depth of field, moody light, golden hour, style by Dan Winters, Russell James, Steve McCurry, centered, extremely detailed, Nikon D850, award winning photography",
    "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k",
    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
    "A photo of beautiful mountain with realistic sunset and blue lake, highly detailed, masterpiece",
]

with gr.Blocks(css="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    gr.DuplicateButton(
        value="Duplicate Space for private use",
        elem_id="duplicate-button",
        visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
    )

    with gr.Group():
        with gr.Row():
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )
            run_button = gr.Button("Run", scale=0)
        result = gr.Gallery(
            label="Generated images", show_label=False, elem_id="gallery", grid=[2]
        )
    with gr.Accordion("Advanced options", open=False):
        seed = gr.Slider(
            label="Seed",
            minimum=0,
            maximum=MAX_SEED,
            step=1,
            value=0,
            randomize=True
        )
        randomize_seed = gr.Checkbox(label="Randomize seed across runs", value=True)
        with gr.Row():
            guidance_scale = gr.Slider(
                label="Guidance scale for base",
                minimum=2,
                maximum=14,
                step=0.1,
                value=8.0,
            )
            num_inference_steps = gr.Slider(
                label="Number of inference steps for base",
                minimum=1,
                maximum=8,
                step=1,
                value=4,
            )
        url = gr.Text(
            label="url",
            value="Null",
            show_label=False,
            placeholder="Null",
            max_lines=1,
            container=False,
            interactive=False,
        )
    
    gr.Examples(
        examples=examples,
        inputs=prompt,
        outputs=result,
        fn=generate,
        cache_examples=CACHE_EXAMPLES,
    )

    gr.on(
        triggers=[
            prompt.submit,
            run_button.click,
        ],
        fn=generate,
        inputs=[
            prompt,
            seed,
            url,
            guidance_scale,
            num_inference_steps,
            randomize_seed
        ],
        outputs=[result, seed, url],
        api_name="run",
    )

if __name__ == "__main__":
    demo.queue(api_open=False)
    delayed_api_request()
    # demo.queue(max_size=20).launch()
    demo.launch()