app.py

import gradio as gr
import numpy as np
import random
import spaces  # [uncomment to use ZeroGPU]
from diffusers import DiffusionPipeline
import torch
import os
from huggingface_hub import login
from openai import OpenAI

# Initialize API keys and login
hf_token = os.getenv("space_token")
openai_api_key = os.getenv("openai_apikey")
login(token=hf_token)

def make_prompt(place):
    client = OpenAI(api_key=openai_api_key)
    messages = [
        {
            "role": "system",
            "content": """userの入力するplace_infoを基に、英単語を、3つ羅列してください。
                        その英単語を基に、男の人の画像生成を行います。英単語の順番は以下の通りです。
                        場所, 行っている動作, 背景の様子

                        ex: The ocean, swimming, with sharks

                        ##output format:
                          place_hoge, moving_hoge, background_hoge"""
        },
        {"role": "user", "content": "place_info: nature"},
        {"role": "assistant", "content": "forest, exploration, there are tigers"},
        {"role": "user", "content": "place_info: " + place}
    ]
    response = client.chat.completions.create(
        model="gpt-4",
        messages=messages,
        temperature=1,
    )
    # Assuming the assistant returns something like "mountains, hiking, clear sky"
    generated_content = response.choices[0].message.content.strip()
    prompt = "Purotan, short brown hair, bright smile, " + generated_content
    return prompt

# Set device and load model
device = "cuda" if torch.cuda.is_available() else "cpu"
model_repo_id = "black-forest-labs/FLUX.1-dev"  # Replace with your model

if torch.cuda.is_available():
    torch_dtype = torch.float16
else:
    torch_dtype = torch.float32

pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
pipe = pipe.to(device)
pipe.load_lora_weights("purotan_1750.safetensors")

# Constants
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024

# Define the image generation function
@spaces.GPU  # [uncomment to use ZeroGPU]
def generate_image(place, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
        
    prompt = make_prompt(place)
    
    generator = torch.Generator().manual_seed(seed)
    
    image = pipe(
        prompt=prompt, 
        guidance_scale=guidance_scale, 
        num_inference_steps=num_inference_steps, 
        width=width, 
        height=height,
        generator=generator
    ).images[0] 
    
    return image, seed

# CSS for styling
css = """
#col-container {
    margin: 0 auto;
    max-width: 640px;
}
"""

# Define the Gradio interface
with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown("""
        # Text-to-Image Gradio Template
        ボタンを押して場所を選択し、画像を生成してください！
        """)
        
        # Place Selection Buttons
        with gr.Row():
            btn_nature = gr.Button("自然")
            btn_cityscape = gr.Button("都市景観")
            btn_fantasy = gr.Button("ファンタジー世界")
            btn_daily = gr.Button("日常生活")
            btn_space = gr.Button("宇宙")
        
        # Display Selected Place
        selected_place_display = gr.Markdown("**選択された場所:** 自然")
        
        # Run Button
        run_button = gr.Button("Run", scale=0)
        
        # Image Output
        result = gr.Image(label="Result", show_label=False)

        # Advanced Settings Accordion
        with gr.Accordion("Advanced Settings", open=False):
            
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )
            
            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
            
            with gr.Row():
                
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=720,  # Adjust based on your model's capabilities
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1280,  # Adjust based on your model's capabilities
                )
            
            with gr.Row():
                
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=0.0,
                    maximum=10.0,
                    step=0.1,
                    value=3.5,  # Adjust based on your model's capabilities
                )
                
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=20,  # Adjust based on your model's capabilities
                )
        
        # Removed the gr.Examples section to fix the ValueError
        # If you wish to add examples, ensure they align with the input components

    # State to keep track of selected place
    selected_place = gr.State("自然")  # Default to "自然"

    # Define functions to set the selected place and update the display
    def set_place(place):
        return place, f"**選択された場所:** {place}"
    
    # Connect buttons to state setter functions using lambda
    btn_nature.click(fn=lambda: set_place("自然"), outputs=[selected_place, selected_place_display])
    btn_cityscape.click(fn=lambda: set_place("都市景観"), outputs=[selected_place, selected_place_display])
    btn_fantasy.click(fn=lambda: set_place("ファンタジー世界"), outputs=[selected_place, selected_place_display])
    btn_daily.click(fn=lambda: set_place("日常生活"), outputs=[selected_place, selected_place_display])
    btn_space.click(fn=lambda: set_place("宇宙"), outputs=[selected_place, selected_place_display])

    # Connect Run button to the image generation function
    run_button.click(
        fn=generate_image,
        inputs=[
            selected_place, 
            seed, 
            randomize_seed, 
            width, 
            height, 
            guidance_scale, 
            num_inference_steps
        ],
        outputs=[result, seed]
    )

demo.queue().launch()