app.py

import os
import torch
import gradio as gr
from diffusers import FluxPipeline
import json
from huggingface_hub import hf_hub_download
import time
from huggingface_hub import login

# Hugging Face hesabında oturum aç
if "HF_TOKEN" in os.environ:
    login(token=os.environ["HF_TOKEN"])
else:
    # İsteğe bağlı: Token yoksa uyarı göster
    print("HF_TOKEN bulunamadı, giriş yapılamıyor")

# Diğer kodlar buradan devam edecek...

# Constants
MODEL_ID = "black-forest-labs/FLUX.1-dev"  # Base model
YOUR_LORA = "anuraj-sisyphus/avatar-loras"  # Your LoRA model
DEFAULT_PROMPT = "a portrait of a person with realistic details, high quality"
DEFAULT_NEG_PROMPT = "low quality, blurry, distorted, deformed features"

# Create a list of available LoRAs
# You can expand this with other compatible LoRAs if desired
LORAS = [
    {
        "name": "Avatar LoRAs",
        "repo_id": "anuraj-sisyphus/avatar-loras",
        "filename": "SLAY1MNSHA.safetensors",  # Update this with the actual filename
        "base_model": "FLUX.1-dev"
    }
]

# Initialize the pipeline
@torch.inference_mode()
def load_model():
    pipe = FluxPipeline.from_pretrained(
        MODEL_ID,
        torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32
    )
    
    if torch.cuda.is_available():
        pipe = pipe.to("cuda")
    
    return pipe

# Generate image function
def generate_image(
    prompt,
    negative_prompt,
    lora_selection,
    lora_scale=0.8,
    guidance_scale=5.0,
    steps=30,
    width=1024,
    height=1024,
    seed=None
):
    # Load model if not already loaded
    global pipe
    if "pipe" not in globals():
        pipe = load_model()
    
    # Set the seed for reproducibility
    if seed is None or seed == 0:
        seed = int(time.time()) % 100000
    generator = torch.Generator(device="cuda" if torch.cuda.is_available() else "cpu").manual_seed(seed)
    
    # Find the selected LoRA details
    selected_lora = None
    for lora in LORAS:
        if lora["name"] == lora_selection:
            selected_lora = lora
            break
    
    if selected_lora:
        # Unload any previous LoRA
        try:
            pipe.unload_lora_weights()
        except:
            pass
            
        # Load the selected LoRA
        pipe.load_lora_weights(
            selected_lora["repo_id"],
            weight_name=selected_lora.get("filename", None)
        )
        
        # Set the LoRA scale
        pipe.fuse_lora(lora_scale=lora_scale)
    
    # Generate the image
    image = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        guidance_scale=guidance_scale,
        num_inference_steps=steps,
        width=width,
        height=height,
        generator=generator
    ).images[0]
    
    return image, seed

# Create the Gradio interface
with gr.Blocks(title="Avatar LoRAs Explorer") as demo:
    gr.Markdown("# Avatar LoRAs Explorer")
    gr.Markdown("Generate images using the Avatar LoRAs model. Adjust settings to customize your results.")
    
    with gr.Row():
        with gr.Column(scale=2):
            prompt = gr.Textbox(
                label="Prompt",
                placeholder="Enter your prompt here...",
                value=DEFAULT_PROMPT,
                lines=3
            )
            negative_prompt = gr.Textbox(
                label="Negative Prompt",
                placeholder="Enter what you don't want to see...",
                value=DEFAULT_NEG_PROMPT,
                lines=2
            )
            
            with gr.Row():
                lora_selection = gr.Dropdown(
                    label="Select LoRA Model",
                    choices=[lora["name"] for lora in LORAS],
                    value=LORAS[0]["name"]
                )
                lora_scale = gr.Slider(
                    label="LoRA Scale",
                    minimum=0.0,
                    maximum=1.5,
                    step=0.05,
                    value=0.8
                )
            
            with gr.Row():
                guidance_scale = gr.Slider(
                    label="Guidance Scale",
                    minimum=1.0,
                    maximum=15.0,
                    step=0.5,
                    value=5.0
                )
                steps = gr.Slider(
                    label="Steps",
                    minimum=10,
                    maximum=100,
                    step=1,
                    value=30
                )
            
            with gr.Row():
                width = gr.Slider(
                    label="Width",
                    minimum=512,
                    maximum=1536,
                    step=64,
                    value=1024
                )
                height = gr.Slider(
                    label="Height",
                    minimum=512,
                    maximum=1536,
                    step=64,
                    value=1024
                )
            
            seed = gr.Number(
                label="Seed (0 for random)",
                value=0,
                precision=0
            )
            
            generate_button = gr.Button("Generate Image", variant="primary")
        
        with gr.Column(scale=2):
            output_image = gr.Image(label="Generated Image", type="pil")
            used_seed = gr.Number(label="Used Seed", value=0, precision=0)
    
    # Setup the button click event
    generate_button.click(
        fn=generate_image,
        inputs=[
            prompt,
            negative_prompt,
            lora_selection,
            lora_scale,
            guidance_scale,
            steps,
            width,
            height,
            seed
        ],
        outputs=[output_image, used_seed]
    )
    
    # Add examples if you have any
    gr.Examples(
        examples=[
            [
                "a portrait photo of a person with blue eyes",
                DEFAULT_NEG_PROMPT,
                LORAS[0]["name"],
                0.8,
                5.0,
                30,
                1024,
                1024,
                42
            ]
        ],
        inputs=[
            prompt,
            negative_prompt,
            lora_selection,
            lora_scale,
            guidance_scale,
            steps,
            width,
            height,
            seed
        ],
        outputs=[output_image, used_seed]
    )

# Launch the app
demo.launch()