App.py

import gradio as gr
import torch
from PIL import Image
import logging
from typing import Optional
import time
from diffusers import StableDiffusionXLImg2ImgPipeline, StableDiffusionXLPipeline

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# ===== CONFIG =====
DEVICE = "cpu"
DTYPE = torch.float32

# ===== PIPELINE MANAGER =====
class PipelineManager:
    def __init__(self):
        self.txt2img_pipe = None
        self.img2img_pipe = None
        self.model_loaded = False
        self.load_lock = False
    
    def load_models(self):
        """Load SDXL models"""
        try:
            logger.info("📥 Loading models...")
            
            self.txt2img_pipe = StableDiffusionXLPipeline.from_pretrained(
                "stabilityai/stable-diffusion-xl-base-1.0",
                torch_dtype=DTYPE,
                use_safetensors=True
            )
            self.txt2img_pipe = self.txt2img_pipe.to(DEVICE)
            self.txt2img_pipe.enable_attention_slicing()
            
            self.img2img_pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(
                "stabilityai/stable-diffusion-xl-base-1.0",
                torch_dtype=DTYPE,
                use_safetensors=True
            )
            self.img2img_pipe = self.img2img_pipe.to(DEVICE)
            self.img2img_pipe.enable_attention_slicing()
            
            self.model_loaded = True
            logger.info("✅ Models loaded!")
            return True
            
        except Exception as e:
            logger.error(f"❌ Error loading models: {e}")
            return False
    
    def initialize(self):
        if self.load_lock:
            return
        self.load_lock = True
        self.load_models()
        self.load_lock = False
    
    def generate_txt2img(
        self,
        prompt: str,
        negative_prompt: str = "",
        num_steps: int = 20,
        guidance: float = 7.5,
        height: int = 768,
        width: int = 768,
        seed: int = -1
    ) -> Image.Image:
        
        if not self.model_loaded:
            raise RuntimeError("Model not loaded")
        
        if seed == -1:
            seed = int(time.time())
        
        generator = torch.Generator(device=DEVICE).manual_seed(seed)
        
        logger.info(f"🎨 Generating: {prompt[:50]}...")
        
        with torch.no_grad():
            image = self.txt2img_pipe(
                prompt=prompt,
                negative_prompt=negative_prompt,
                num_inference_steps=num_steps,
                guidance_scale=guidance,
                height=height,
                width=width,
                generator=generator
            ).images[0]
        
        return image
    
    def generate_img2img(
        self,
        prompt: str,
        image: Image.Image,
        negative_prompt: str = "",
        num_steps: int = 20,
        guidance: float = 7.5,
        strength: float = 0.8,
        seed: int = -1
    ) -> Image.Image:
        
        if not self.model_loaded:
            raise RuntimeError("Model not loaded")
        
        if seed == -1:
            seed = int(time.time())
        
        generator = torch.Generator(device=DEVICE).manual_seed(seed)
        image = image.resize((768, 768), Image.Resampling.LANCZOS)
        
        logger.info(f"🖼️ Transforming: {prompt[:50]}...")
        
        with torch.no_grad():
            image = self.img2img_pipe(
                prompt=prompt,
                image=image,
                negative_prompt=negative_prompt,
                num_inference_steps=num_steps,
                guidance_scale=guidance,
                strength=strength,
                generator=generator
            ).images[0]
        
        return image

pipeline_manager = PipelineManager()

# ===== UI FUNCTIONS =====

def txt2img(prompt, neg_prompt, steps, guidance, height, width, seed):
    try:
        if not pipeline_manager.model_loaded:
            return None, "❌ Model loading..."
        image = pipeline_manager.generate_txt2img(prompt, neg_prompt, steps, guidance, height, width, seed)
        return image, "✅ Done!"
    except Exception as e:
        return None, f"❌ {str(e)}"

def img2img(prompt, input_image, neg_prompt, steps, guidance, strength, seed):
    try:
        if input_image is None:
            return None, "❌ Upload image first"
        if not pipeline_manager.model_loaded:
            return None, "❌ Model loading..."
        image = pipeline_manager.generate_img2img(prompt, input_image, neg_prompt, steps, guidance, strength, seed)
        return image, "✅ Done!"
    except Exception as e:
        return None, f"❌ {str(e)}"

# ===== GRADIO UI =====

with gr.Blocks(title="FLUX Generator") as demo:
    gr.Markdown("# 🎨 FLUX - Image Generator")
    
    with gr.Tabs():
        
        with gr.Tab("📝 Text-to-Image"):
            with gr.Row():
                with gr.Column():
                    prompt = gr.Textbox(label="Prompt", lines=3, placeholder="Describe image...")
                    neg_prompt = gr.Textbox(label="Negative", lines=2, placeholder="What to avoid...")
                    
                    with gr.Row():
                        height = gr.Slider(256, 1024, 768, 64, label="Height")
                        width = gr.Slider(256, 1024, 768, 64, label="Width")
                    
                    with gr.Row():
                        steps = gr.Slider(1, 50, 20, 1, label="Steps")
                        guidance = gr.Slider(1, 15, 7.5, 0.5, label="Guidance")
                    
                    seed = gr.Number(-1, label="Seed (-1=random)", precision=0)
                    btn = gr.Button("🎨 Generate", variant="primary", size="lg")
                
                with gr.Column():
                    output = gr.Image(label="Output")
                    status = gr.Textbox(interactive=False, label="Status")
            
            btn.click(txt2img, [prompt, neg_prompt, steps, guidance, height, width, seed], [output, status])
        
        with gr.Tab("🖼️ Image-to-Image"):
            with gr.Row():
                with gr.Column():
                    img_input = gr.Image(label="Input Image", type="pil")
                    prompt2 = gr.Textbox(label="Prompt", lines=3, placeholder="Transform to...")
                    neg_prompt2 = gr.Textbox(label="Negative", lines=2)
                    
                    with gr.Row():
                        steps2 = gr.Slider(1, 50, 20, 1, label="Steps")
                        guidance2 = gr.Slider(1, 15, 7.5, 0.5, label="Guidance")
                    
                    strength = gr.Slider(0, 1, 0.8, 0.05, label="Strength")
                    seed2 = gr.Number(-1, label="Seed (-1=random)", precision=0)
                    btn2 = gr.Button("🖼️ Generate", variant="primary", size="lg")
                
                with gr.Column():
                    output2 = gr.Image(label="Output")
                    status2 = gr.Textbox(interactive=False, label="Status")
            
            btn2.click(img2img, [prompt2, img_input, neg_prompt2, steps2, guidance2, strength, seed2], [output2, status2])

def on_load():
    logger.info("🚀 Loading pipeline...")
    pipeline_manager.initialize()
    if pipeline_manager.model_loaded:
        return "✅ Ready!"
    return "⏳ Loading models..."

gr.on_load(on_load)

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860, share=True)