app.py

import gradio as gr
import numpy as np
import random
import spaces
import torch
from diffusers import  DiffusionPipeline, FlowMatchEulerDiscreteScheduler, AutoencoderTiny, AutoencoderKL, StableDiffusionUpscalePipeline
from transformers import CLIPTextModel, CLIPTokenizer,T5EncoderModel, T5TokenizerFast
from live_preview_helpers import calculate_shift, retrieve_timesteps, flux_pipe_call_that_returns_an_iterable_of_images
from huggingface_hub import hf_hub_download
import os
import requests

dtype = torch.bfloat16
device = "cuda" if torch.cuda.is_available() else "cpu"

taef1 = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype).to(device)
good_vae = AutoencoderKL.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="vae", torch_dtype=dtype).to(device)
pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=dtype, vae=taef1).to(device)

# Performance optimizations
if hasattr(pipe, "enable_model_cpu_offload"):
    pipe.enable_model_cpu_offload()
if hasattr(pipe, "enable_attention_slicing"):
    pipe.enable_attention_slicing(1)
if hasattr(pipe, "enable_vae_slicing"):
    pipe.enable_vae_slicing()
if hasattr(pipe, "enable_vae_tiling"):
    pipe.enable_vae_tiling()

# Compile transformer for faster inference (if supported)
try:
    pipe.transformer = torch.compile(pipe.transformer, mode="reduce-overhead", fullgraph=True)
    print("✓ Transformer compiled for faster inference")
except Exception as e:
    print(f"Warning: Could not compile transformer: {e}")

# Load upscaler pipeline with optimizations
upscaler = StableDiffusionUpscalePipeline.from_pretrained("stabilityai/stable-diffusion-x4-upscaler", torch_dtype=dtype).to(device)
if hasattr(upscaler, "enable_model_cpu_offload"):
    upscaler.enable_model_cpu_offload()
if hasattr(upscaler, "enable_attention_slicing"):
    upscaler.enable_attention_slicing(1)
if hasattr(upscaler, "enable_vae_slicing"):
    upscaler.enable_vae_slicing()

# Available LoRAs
LORAS = {
    "None": None,
    "AntiBlur": "Shakker-Labs/FLUX.1-dev-LoRA-AntiBlur",
    "Add Details": "Shakker-Labs/FLUX.1-dev-LoRA-add-details",
    "Ultra Realism": "https://huggingface.co/its-magick/merlin-test-loras/resolve/main/Canopus-LoRA-Flux-UltraRealism.safetensors",
    "Face Realism": "https://huggingface.co/its-magick/merlin-test-loras/resolve/main/Canopus-LoRA-Flux-FaceRealism.safetensors"
}

# Store loaded LoRA paths
loaded_loras = {}

def download_lora_from_url(url, filename):
    """Download LoRA file from direct URL"""
    if not os.path.exists(filename):
        print(f"Downloading {filename}...")
        response = requests.get(url)
        with open(filename, 'wb') as f:
            f.write(response.content)
        print(f"Downloaded {filename}")
    return filename

def preload_and_apply_all_loras():
    """Download and apply all LoRAs simultaneously at startup"""
    global loaded_loras
    
    print("Downloading and applying all LoRAs...")
    
    for lora_name, lora_path in LORAS.items():
        if lora_name == "None" or lora_path is None:
            continue
            
        # Handle direct URL downloads
        if lora_path.startswith('http'):
            filename = f"{lora_name.lower().replace(' ', '_')}_lora.safetensors"
            lora_path = download_lora_from_url(lora_path, filename)
        
        loaded_loras[lora_name] = lora_path
        print(f"Downloaded {lora_name}")
        
        # Apply each LoRA with optimal scaling
        try:
            optimal_scale = get_optimal_lora_scale(lora_name)
            pipe.load_lora_weights(lora_path, adapter_name=lora_name.lower().replace(' ', '_'))
            print(f"Applied {lora_name} with scale {optimal_scale}")
        except Exception as e:
            print(f"Failed to apply {lora_name}: {e}")
    
    print(f"All {len(loaded_loras)} LoRAs downloaded and applied!")

def get_optimal_lora_scale(lora_name):
    """Return optimal LoRA scale based on LoRA type for better quality/speed balance"""
    lora_scales = {
        "AntiBlur": 0.8,  # Slightly lower for better balance
        "Add Details": 1.2,  # Higher for more detail enhancement
        "Ultra Realism": 0.9,  # Balanced for realism
        "Face Realism": 1.1,  # Optimized for facial features
    }
    return lora_scales.get(lora_name, 1.0)

# Download and apply all LoRAs at startup
preload_and_apply_all_loras()

torch.cuda.empty_cache()

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(pipe)

@spaces.GPU(duration=75)
def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=3.5, num_inference_steps=28, enable_upscale=False, progress=gr.Progress(track_tqdm=True)):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator().manual_seed(seed)
    
    # All LoRAs are already loaded and active
    
    try:
        final_img = None
        for img in pipe.flux_pipe_call_that_returns_an_iterable_of_images(
                prompt=prompt,
                guidance_scale=guidance_scale,
                num_inference_steps=num_inference_steps,
                width=width,
                height=height,
                generator=generator,
                output_type="pil",
                good_vae=good_vae,
            ):
                final_img = img
                yield img, seed
        
        # Apply upscaling if enabled with optimized settings
        if enable_upscale and final_img is not None:
            try:
                # Use fewer steps for faster upscaling with minimal quality loss
                upscaled_img = upscaler(
                    prompt=prompt,
                    image=final_img,
                    num_inference_steps=15,  # Reduced from 20 for speed
                    guidance_scale=6.0,  # Slightly lower for faster convergence
                    generator=generator,
                ).images[0]
                yield upscaled_img, seed
            except Exception as e:
                print(f"Error during upscaling: {e}")
                yield final_img, seed
                
    except Exception as e:
        print(f"Error during generation: {e}")
        # Fallback to basic generation
        img = pipe(
            prompt=prompt,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            width=width,
            height=height,
            generator=generator,
        ).images[0]
        
        # Apply upscaling if enabled
        if enable_upscale:
            try:
                img = upscaler(
                    prompt=prompt,
                    image=img,
                    num_inference_steps=20,
                    guidance_scale=7.5,
                    generator=generator,
                ).images[0]
            except Exception as e:
                print(f"Error during upscaling: {e}")
                
        yield img, seed
    
examples = [
    "a tiny astronaut hatching from an egg on the moon",
    "a cat holding a sign that says hello world",
    "an anime illustration of a wiener schnitzel",
]

css="""
#col-container {
    margin: 0 auto;
    max-width: 520px;
}
"""

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""# FLUX.1 [dev]
12B param rectified flow transformer guidance-distilled from [FLUX.1 [pro]](https://blackforestlabs.ai/)  
[[non-commercial license](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md)] [[blog](https://blackforestlabs.ai/announcing-black-forest-labs/)] [[model](https://huggingface.co/black-forest-labs/FLUX.1-dev)]
        """)
        
        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.Image(label="Result", show_label=False)
        
        with gr.Accordion("Advanced Settings", open=False):
            
            gr.Markdown("**LoRAs Active:** All LoRAs are loaded and active simultaneously")
            
            enable_upscale = gr.Checkbox(
                label="Enable 4x Upscaling",
                value=False,
                info="Upscale final image using Stable Diffusion 4x upscaler"
            )
            
            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=1024,
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
            
            with gr.Row():

                guidance_scale = gr.Slider(
                    label="Guidance Scale",
                    minimum=1,
                    maximum=15,
                    step=0.1,
                    value=3.5,
                    info="Lower values = faster generation, higher values = more prompt adherence"
                )
  
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=4,
                    maximum=50,
                    step=1,
                    value=20,
                    info="Lower values = faster generation, higher values = better quality"
                )
        
        gr.Examples(
            examples = examples,
            fn = infer,
            inputs = [prompt],
            outputs = [result, seed],
            cache_examples="lazy"
        )

    gr.on(
        triggers=[run_button.click, prompt.submit],
        fn = infer,
        inputs = [prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, enable_upscale],
        outputs = [result, seed]
    )

demo.launch(share=True)