import torch
from diffusers import FluxPipeline
from transformers import pipeline
import gradio as gr
import spaces


device=torch.device('cuda')

pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.to("cuda")
 




# Load the NSFW classifier
image_classifier = pipeline("image-classification", model="Falconsai/nsfw_image_detection",device=device)
#text_classifier = pipeline("text-classification", model="eliasalbouzidi/distilbert-nsfw-text-classifier",device=device)

NSFW_THRESHOLD = 0.3

# Define the function to generate the sketch
@spaces.GPU
def generate_sketch(prompt,style, num_inference_steps, guidance_scale):
    # Classify the text for NSFW content
    #text_classification = text_classifier(prompt)
    #print(text_classification)
    
    # Check the classification results
    #for result in text_classification:
    #    if result['label'] == 'nsfw' and result['score'] > NSFW_THRESHOLD: 
    #        return gr.update(visible=False),gr.Text(value="Inappropriate prompt detected. Please try another prompt.")
    print(prompt)
    

    match style:
        case 'shou_xin':
            prompt = "shou_xin, " + prompt
            pipe.load_lora_weights("Datou1111/shou_xin", weight_name="shou_xin.safetensors")
        case 'sketched':
            prompt = "sketched style, " + prompt
            pipe.load_lora_weights("Shakker-Labs/FLUX.1-dev-LoRA-Children-Simple-Sketch", weight_name="FLUX-dev-lora-children-simple-sketch.safetensors")
        case 'sketch_paint':
            prompt = "Sketch paint, " + prompt
            pipe.load_lora_weights("strangerzonehf/Sketch-Paint", weight_name="Sketch-Paint.safetensors")
        case 'sketch_sized':
            prompt = "Sketch Sized, " + prompt
            pipe.load_lora_weights("strangerzonehf/Flux-Sketch-Sized-LoRA", weight_name="Flux-Sketch-Sized-LoRA.safetensors")
        case _:
            prompt = "shou_xin, " + prompt
            pipe.load_lora_weights("Datou1111/shou_xin", weight_name="shou_xin.safetensors")

    
    pipe.fuse_lora(lora_scale=1.5)
    
        
    image = pipe("sketched style, " + prompt, 
                 num_inference_steps=num_inference_steps, 
                 guidance_scale=guidance_scale,
                ).images[0]
    

    # Classify the image for NSFW content
    image_classification = image_classifier(image)

    print(image_classification)

    # Check the classification results
    for result in image_classification:
        if result['label'] == 'nsfw' and result['score'] > NSFW_THRESHOLD: 
           return None,"Inappropriate content detected. Please try another prompt." #return gr.update(visible=False),gr.Text(value="Inappropriate content detected. Please try another prompt.")
    
    image_path = "generated_sketch.png"
    
    image.save(image_path)
    return image_path,None #gr.Image(value=image_path), gr.update(visible=False)

# Gradio interface with sliders for num_inference_steps and guidance_scale
interface = gr.Interface(
    fn=generate_sketch,
    inputs=[
        "text",  # Prompt input
        gr.Dropdown(
        ["sketched", "shou_xin","Sketch-Paint","Flux-Sketch-Sized-LoRA"],value='sketched', label="Style"
        ),
        gr.Slider(5, 50, value=24, step=1, label="Number of Inference Steps"),  # Slider for num_inference_steps
        gr.Slider(1.0, 10.0, value=3.5, step=0.1, label="Guidance Scale")  # Slider for guidance_scale
    ],
    outputs=[
        gr.Image(label="Generated Sketch"), 
        gr.Textbox(label="Message")  
    ],
    title="Kids Sketch Generator",
    description="Enter a text prompt and generate a fun sketch for kids with customizable inference steps and guidance scale."
)

# Launch the app
interface.launch()