app.py

import gradio as gr
import os
import time
import requests
from PIL import Image


# ux format
tryon_css="""
#col-garment {
    margin: 0 auto;
    max-width: 420px;
}
#garm_img {
    aspect-ratio: 3 / 4;
    width: 100%;
    max-height: 560px;
    object-fit: contain;
}
#col-person {
    margin: 0 auto;
    max-width: 420px;
}
#person_img {
    aspect-ratio: 3 / 4;
    width: 100%;
    max-height: 560px;
    object-fit: contain;
}
#col-result {
    margin: 0 auto;
    max-width: 420px;
}
#result_img {
    aspect-ratio: 3 / 4;
    width: 100%;
    max-height: 560px;
    object-fit: contain;
}
#col-examples {
    margin: 0 auto;
    max-width: 1000px;
}
#col-examples img {
    aspect-ratio: 3 / 4;
    object-fit: contain;
}
#button {
    background-color: #A47764;
    color: white;
}
"""

# assets loading
example_path = os.path.join(os.path.dirname(__file__), 'data')

garm_list = os.listdir(os.path.join(example_path,"garment"))
garm_list_path = [os.path.join(example_path, "garment", garm) for garm in garm_list]

person_list = os.listdir(os.path.join(example_path,"person"))
person_list_path = [os.path.join(example_path, "person", person) for person in person_list]

garm_img_category_mapping = {os.path.basename(garm_file): os.path.basename(garm_file).split("_")[2].capitalize() for garm_file in garm_list_path}


def load_header(header_file):
    with open(header_file, 'r', encoding='utf-8') as f:
        content = f.read()
    return content

def preprocess_img(img_path, max_size=1024):
    if img_path is None:
        return None
    img = Image.open(img_path)
    if max(img.size) > max_size:
        img.thumbnail((max_size, max_size))
        img.save(img_path)
    return img_path

def update_category(selected_garm_file):
    selected_category = garm_img_category_mapping.get(os.path.basename(selected_garm_file), "Fullbody")
    return gr.update(value=selected_category)

def call_tryon_api(person_file, garm_file, category, model_type='SD_V1'):
    tryon_url = os.environ['API_ENDPOINT'] + "/tryon/v1"
    payload = {'garment_type': category, 'model_type': model_type, 'repaint_other_garment': 'false'}
    files = {
        'image_garment_file': open(garm_file, 'rb'),
        'image_model_file': open(person_file, 'rb'),   
        }
    headers = {
        'x-api-key': os.environ['API_KEY']
    }
    
    try:
        response = requests.post(tryon_url, headers=headers, data=payload, files=files)
        if response.ok:
            data = response.json()
            return data['job_id'], data['status']
        else:
            print(response.content)
    except Exception as e:
        print(f"call tryon api error: {e}")
    
    # if the API call fails, return pop up error
    raise gr.Error("Over heated, please try again later")

def get_tryon_result(job_id):
    result_url = os.environ['API_ENDPOINT'] + "/requests/v1" + f"?job_id={job_id}"
    headers = {
        'x-api-key': os.environ['API_KEY']
    }
    
    try:
        response = requests.get(result_url, headers=headers)
    
        if response.ok:
            data = response.json()
            if data["status"] == "completed":
                image_url = data['output'][0]['image_url']
                return image_url, data['status']
            else:
                return None, data['status']
    except Exception as e:
        print(f"get tryon result error: {e}")
        return None, None

def run_turbo(person_img, garm_img, category="Top"):
    if person_img is None or garm_img is None:
        gr.Warning("input image is missing")
        return None, "No input image"
    
    info = "" # placeholder for now

    job_id, status = call_tryon_api(person_img, garm_img, category, model_type= os.environ['MODEL_TYPE'])
    
    time.sleep(8) # wait before fetching the result
    
    # check the status of the job
    max_retry = 40  # 40x1.5s = 60s timeout for sinlge job run
    while status not in ["completed", "failed"]:
        try:
            result_image_url, status = get_tryon_result(job_id)
            if result_image_url is not None:
                return result_image_url, info
        except:
            pass
        time.sleep(1.5)  # Wait before retrying

    gr.Warning("Over heated, please try again later")
    return None, info

with gr.Blocks(css=tryon_css) as Huhu_Turbo:
    gr.HTML(load_header("data/header.html"))
    with gr.Row():
        with gr.Column(elem_id = "col-garment"):
            gr.HTML("""
            <div style="display: flex; justify-content: center; align-items: center; text-align: center; font-size: 20px;">
                <div>
                Upload your garment image 🧥
                </div>
            </div>
            """)
        with gr.Column(elem_id = "col-person"):
            gr.HTML("""
            <div style="display: flex; justify-content: center; align-items: center; text-align: center; font-size: 20px;">
                <div>
                Select a model image 🧍
                </div>
            </div>
            """)
        with gr.Column(elem_id = "col-result"):
            gr.HTML("""
            <div style="display: flex; justify-content: center; align-items: center; text-align: center; font-size: 20px;">
                <div>
                “RUN” to get results 🪄
                </div>
            </div>
            """)
    with gr.Row():
        with gr.Column(elem_id = "col-garment"):
            garm_img = gr.Image(label="Garment image", sources='upload', type="filepath", elem_id="garm_img")
            category = gr.Dropdown(label="Garment type", choices=['Top', 'Bottom', 'Fullbody'],  value="Top")
            garm_example = gr.Examples(
                inputs=garm_img,
                examples_per_page=10,
                examples=garm_list_path,
                cache_examples=False 
            )
        with gr.Column(elem_id = "col-person"):
            person_img = gr.Image(label="Person image", sources='upload', type="filepath", elem_id="person_img")
            person_example = gr.Examples(
                inputs=person_img,
                examples_per_page=10,
                examples=person_list_path
            )
        with gr.Column(elem_id = "col-result"):
            result_img = gr.Image(label="Result", show_share_button=False, elem_id="result_img")
            with gr.Row():
                result_info = gr.Text(label="Tryon inference runtime", visible=False)
            generate_button = gr.Button(value="RUN", elem_id="button")

    garm_example.load_input_event.then(
        fn=update_category,
        inputs=[garm_img],
        outputs=[category]
    )
    
    garm_img.change(fn=preprocess_img, inputs=[garm_img], outputs=[garm_img])
    person_img.change(fn=preprocess_img, inputs=[person_img], outputs=[person_img])

    generate_button.click(fn=run_turbo, inputs=[person_img, garm_img, category], outputs=[result_img, result_info], api_name=False, concurrency_limit=30)

    gr.HTML(load_header("data/note.html"))
    
    with gr.Column(elem_id = "col-examples"):
        gr.HTML("""
        <div style="display: flex; justify-content: center; align-items: center; text-align: center; font-size: 20px;">
            <div> </div>
            <br>
            <div>
            Huhu Try-on Turbo examples in pairs of garment and model images
            </div>
        </div>
        """)
        show_case = gr.Examples(
            examples=[
                ["data/examples/person_example_1.png", "data/examples/garment_example_1.png", "Top", "data/examples/result_example_1.png"],
                ["data/examples/person_example_2.png", "data/examples/garment_example_2.png", "Top", "data/examples/result_example_2.png"],
                ["data/examples/person_example_3.png", "data/examples/garment_example_3.png", "Top", "data/examples/result_example_3.png"],
                ["data/examples/person_example_4.png", "data/examples/garment_example_4.png", "Fullbody", "data/examples/result_example_4.png"],
                ["data/examples/person_example_5.png", "data/examples/garment_example_5.png", "Top", "data/examples/result_example_5.png"],
            ],
            inputs=[person_img, garm_img, category, result_img],
            label=None
        )

Huhu_Turbo.queue(api_open=False).launch(show_api=False)