app.py

import gradio as gr
import requests
import io
import random
import os
from PIL import Image
from huggingface_hub import InferenceClient
from deep_translator import GoogleTranslator
from gradio_client import Client
import logging
from datetime import datetime
token = os.getenv('HF_READ_TOKEN')
from logger import log


# os.makedirs('assets', exist_ok=True)
if not os.path.exists('icon.png'):
    os.system("wget -O icon.png https://huggingface.co/spaces/K00B404/FLUX.1-Dev-Serverless-darn-enhanced-prompt/resolve/main/edge.png")
API_URL_DEV = "https://api-inference.huggingface.co/models/black-forest-labs/FLUX.1-dev"
API_URL = "https://api-inference.huggingface.co/models/black-forest-labs/FLUX.1-schnell"

chatter="K00B404/transcript_image_generator"
max_tokens_enhance_bot=128
# Initialize the API client for the chatbot
chatbot_client = Client(chatter)

timeout = 100


def check_ubuse(prompt,word_list=["little girl"]):
    for word in word_list:
        if word in prompt:
            print(f"Abuse! prompt {prompt} wiped!")
            return "None"
    return prompt
    
def enhance_prompt(prompt, model="mistralai/Mistral-7B-Instruct-v0.1", style="photo-realistic"):
    
    client = Client("K00B404/Mistral-Nemo-custom")
    
    system_prompt=f"""
    You are a image generation prompt enhancer specialized in the {style} style. 
    You must respond only with the enhanced version of the users input prompt
    Remember, image generation models can be stimulated by refering to camera 'effect' in the prompt like :4k ,award winning, super details, 35mm lens, hd
    """
    user_message=f"###input image generation prompt### {prompt}"
   
    result = client.predict(
    		system_prompt=system_prompt,
    		user_message=user_message,
    		max_tokens=256,
    		model_id=model,# "mistralai/Mistral-Nemo-Instruct-2407",
    		api_name="/predict"
    )
    return result
    
    # The output value that appears in the "Response" Textbox component.
    """result = client.predict(
        system_prompt=system_prompt,#"You are a image generation prompt enhancer and must respond only with the enhanced version of the users input prompt",
        user_message=user_message,
        max_tokens=500,
        api_name="/predict"
    )
    return result
    """


def enhance_prompt_v2(prompt, model="mistralai/Mistral-Nemo-Instruct-2407", style="photo-realistic"):
    
    client = Client("K00B404/Mistral-Nemo-custom")
    
    system_prompt=f"""
    You are a image generation prompt enhancer specialized in the {style} style. 
    You must respond only with the enhanced version of the users input prompt
    Remember, image generation models can be stimulated by refering to camera 'effect' in the prompt like :4k ,award winning, super details, 35mm lens, hd
    """
    user_message=f"###input image generation prompt### {prompt}"
   
    result = client.predict(
    		system_prompt=system_prompt,
    		user_message=user_message,
    		max_tokens=max_tokens_enhance_bot,
    		model_id=model,
    		api_name="/predict"
    )
    return result

    
def mistral_nemo_call(prompt, API_TOKEN, model="mistralai/Mistral-Nemo-Instruct-2407", style="photo-realistic"):
    
    client = InferenceClient(api_key=API_TOKEN)
    system_prompt=f"""
    You are a image generation prompt enhancer specialized in the {style} style. 
    You must respond only with the enhanced version of the users input prompt
    Remember, image generation models can be stimulated by refering to camera 'effect' in the prompt like :4k ,award winning, super details, 35mm lens, hd
    """
    
    response = ""
    for message in client.chat_completion(
        model=model,
        messages=[{"role": "system", "content": system_prompt},
                  {"role": "user", "content": prompt}
                 ],
        max_tokens=max_tokens_enhance_bot,
        stream=True,
    ):
        response += message.choices[0].delta.content
    return response



def chat_with_persona(message, history, system_message, max_tokens, temperature, top_p):
    """Function to interact with the chatbot API using the generated persona"""
    try:
        # Call the API with the current message and system prompt (persona)
        response = chatbot_client.predict(
            message=message,
            system_message=system_message,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            api_name="/chat"
        )
        return response
    except Exception as e:
        return f"Error communicating with the chatbot API: {str(e)}"
        

def query(prompt, is_negative=False, steps=30, cfg_scale=7, sampler="DPM++ 2M Karras", seed=-1, 
          strength=0.7, huggingface_api_key=None, use_dev=False,
          enhance_prompt_style="generic", enhance_prompt_option=False, 
          nemo_enhance_prompt_style="generic", use_mistral_nemo=False):

    # Log the request (WITHOUT storing API keys)

    DATA=f"{prompt}|{is_negative}|{steps}|{cfg_scale}|{sampler}|{seed}|{strength}|{use_dev}|{enhance_prompt_style}|{enhance_prompt_option}|{nemo_enhance_prompt_style}|{use_mistral_nemo}"
    log(file_name="FluxCapacitor_log", data=f"{DATA}")


    # Determine API URL
    api_url = API_URL_DEV if use_dev else API_URL

    # API key handling
    is_api_call = huggingface_api_key is not None
    if is_api_call:
        API_TOKEN = os.getenv("HF_READ_TOKEN")  # Use env var in GUI mode
    else:
        if huggingface_api_key == "":
            raise gr.Error("API key is required for API calls.")
        API_TOKEN = huggingface_api_key
    
    headers = {"Authorization": f"Bearer {API_TOKEN}"} 

    if not prompt:  # Simplified check
        return None, None, None

    key = random.randint(0, 999)
    prompt = check_ubuse(prompt)
    print(f'\033[1mGeneration {key} translation:\033[0m {prompt}')

    original_prompt = prompt
    if enhance_prompt_option:
        style=enhance_prompt_style
        system_prompt=f"""
    You are a image generation prompt enhancer specialized in the {style} style. 
    You must respond only with the enhanced version of the users input prompt
    Remember, image generation models can be stimulated by refering to camera 'effect' in the prompt like :4k ,award winning, super details, 35mm lens, hd
    """
        prompt = chat_with_persona(message=prompt, history=[], system_message=system_prompt, max_tokens=max_tokens_enhance_bot, temperature=0.1, top_p=0.97)
        #prompt = enhance_prompt_v2(prompt, style=enhance_prompt_style)
        print(f'\033[1mGeneration {key} enhanced prompt:\033[0m {prompt}')
        
    if use_mistral_nemo:
        style=nemo_enhance_prompt_style
        system_prompt=f"""
    You are a image generation prompt enhancer specialized in the {style} style. 
    You must respond only with the enhanced version of the users input prompt
    Remember, image generation models can be stimulated by refering to camera 'effect' in the prompt like :4k ,award winning, super details, 35mm lens, hd
    """
        prompt = chat_with_persona(message=prompt, history=[], system_message=system_prompt, max_tokens=max_tokens_enhance_bot, temperature=0.1, top_p=0.97)
        #prompt = mistral_nemo_call(prompt, API_TOKEN=API_TOKEN, style=nemo_enhance_prompt_style)
        print(f'\033[1mGeneration {key} Mistral-Nemo prompt:\033[0m {prompt}')
        
    final_prompt = f"{prompt} | ultra detail, ultra elaboration, ultra quality, perfect."
    print(f'\033[1mGeneration {key}:\033[0m {final_prompt}')

    # Ensure seed is always positive
    if seed == -1:
        seed = random.randint(1, 1000000000)

    payload = {
        "inputs": final_prompt,
        "is_negative": is_negative,
        "steps": steps,
        "cfg_scale": cfg_scale,
        "seed": seed,
        "strength": strength
    }

    response = requests.post(api_url, headers=headers, json=payload, timeout=timeout)
    if response.status_code != 200:
        print(f"Error: Failed to get image. Response status: {response.status_code}")
        print(f"Response content: {response.text}")
        if response.status_code == 503:
            raise gr.Error(f"{response.status_code} : The model is being loaded")
        raise gr.Error(f"{response.status_code}")
    
    try:
        image_bytes = response.content
        image = Image.open(io.BytesIO(image_bytes))
        print(f'\033[1mGeneration {key} completed!\033[0m ({final_prompt})')

        # Save the image and return path
        output_path = f"./output_{key}.png"
        image.save(output_path)
        
        return output_path, seed, prompt if enhance_prompt_option else original_prompt
    except Exception as e:
        print(f"Error when trying to open the image: {e}")
        return None, None, None
  
title_html="""
    <center>
        <div id="title-container">
            <h1 id="title-text">FLUX Capacitor</h1>
        </div>
    </center>
"""

css = """
.gradio-container {
    background: url(https://huggingface.co/spaces/K00B404/FLUX.1-Dev-Serverless-darn-enhanced-prompt/resolve/main/edge.png);
    background-size: 900px 1880px;
    background-repeat: no-repeat;
    background-position: center;
    background-attachment: fixed;
    color:#000;
}
.dark\:bg-gray-950:is(.dark *) {
  --tw-bg-opacity: 1;
  background-color: rgb(0, 17, 0);
}

.gradio-container-4-41-0 .prose :last-child {
  margin-top: 8px !important;
}
.gradio-container-4-41-0 .prose :last-child {
  margin-bottom: -7px !important;
}
.dark {
    --button-primary-background-fill: #000;
    --button-primary-background-fill-hover: #00000070;
    --background-fill-primary: #000;
    --background-fill-secondary: #000;
}
.hide-container {
    margin-top;-2px;
}
#app-container3 {
    background-color: rgba(255, 255, 255, 0.001);  /* Corrected to make semi-transparent */
    max-width: 1600px;
    margin-left: auto;
    margin-right: auto;
    margin-bottom: 10px;
    border-radius: 125px;
    box-shadow: 0 0 10px rgba(0,0,0,0.1); /* Adjusted shadow opacity */
}
#app-container {
    background-color: rgba(255, 255, 255, 0.001);  /* Semi-transparent background */
    max-width: 600px;
    margin: 0 auto;  /* Center horizontally */
    padding-bottom: 10px;
    border-radius: 25px;
    box-shadow: 0 0 10px rgba(0, 0, 0, 0.1); /* Adjusted shadow opacity */
}
.panel-container {
    background-image: url('your-neon-border-image.png');
    background-size: 100% 100%; /* Adjust the size to cover the container */
    background-repeat: no-repeat;
    background-position: center;
}
#title-container {
    display: flex;
    align-items: center
    margin-bottom:10px;
    justify-content: center;
}

#title-icon {
    width: 32px;
    height: auto;
    margin-right: 10px;
}

#title-text {
    font-size: 30px;
    font-weight: bold;
    color: #000;
}
:root {
  --panel-size: 300px;
  --border-width: 4px;
  --glow-blur: 15px;
}

body {
  background-color: #000;
  display: flex;
  justify-content: center;
  align-items: center;
  min-height: 100vh;
  margin: 0;
}

.neon-panel {
  width: var(--panel-size);
  height: var(--panel-size);
  background-color: #000;
  position: relative;
  border-radius: 20px;
  overflow: hidden;
}

.neon-panel::before,
.neon-panel::after {
  content: '';
  position: absolute;
  left: -2px;
  top: -2px;
  background: linear-gradient(
    124deg,
    #ff2400, #e81d1d, #e8b71d, #e3e81d, #1de840,
    #1ddde8, #2b1de8, #000, #dd00f3
  );
  background-size: 300% 300%;
  width: calc(100% + 4px);
  height: calc(100% + 4px);
  z-index: -1;
  animation: moveGradient 10s ease infinite;
}

.neon-panel::after {
  filter: blur(var(--glow-blur));
}

.neon-panel-content {
  position: absolute;
  top: var(--border-width);
  left: var(--border-width);
  right: var(--border-width);
  bottom: var(--border-width);
  background-color: #000;
  border-radius: 16px;
  z-index: 1;
}

@keyframes moveGradient {
  0% { background-position: 0% 50%; }
  50% { background-position: 100% 50%; }
  100% { background-position: 0% 50%; }
}

@media (max-width: 768px) {
  :root {
    --panel-size: 250px;
    --glow-blur: 10px;
  }
}

@media (prefers-reduced-motion: reduce) {
  .neon-panel::before,
  .neon-panel::after {
    animation: none;
  }
}
"""
js_script = """
<script>
function detectWindowChange() {
    let lastX = window.screenX;
    let lastY = window.screenY;
    let lastW = window.innerWidth;
    let lastH = window.innerHeight;
    
    setInterval(async function() {
        if (window.screenX !== lastX || window.screenY !== lastY || 
            window.innerWidth !== lastW || window.innerHeight !== lastH) {
            
            lastX = window.screenX;
            lastY = window.screenY;
            lastW = window.innerWidth;
            lastH = window.innerHeight;
            
            let response = await fetch("/get_bounds");
            let data = await response.json();
            updateNeonGlow(data.x, data.y, data.w, data.h);
        }
    }, 500);
}

detectWindowChange();
</script>
"""

with gr.Blocks( css=css) as app:
    gr.HTML(title_html+ js_script) # title html 
    with gr.Column(elem_id="app-container"):
        with gr.Row():
            with gr.Column(elem_id="prompt-container"):
                with gr.Row():
                    text_prompt = gr.Textbox(label="Prompt", placeholder="Enter a prompt here", lines=2, elem_id="prompt-text-input")
                with gr.Row():
                    with gr.Accordion("Advanced Settings", open=False):
                        negative_prompt = gr.Textbox(label="Negative Prompt", placeholder="What should not be in the image", value="(deformed, distorted, disfigured), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, misspellings, typos", lines=3, elem_id="negative-prompt-text-input")
                        steps = gr.Slider(label="Sampling steps", value=35, minimum=1, maximum=100, step=1)
                        cfg = gr.Slider(label="CFG Scale", value=7, minimum=1, maximum=20, step=1)
                        method = gr.Radio(label="Sampling method", value="DPM++ 2M Karras", choices=["DPM++ 2M Karras", "DPM++ SDE Karras", "Euler", "Euler a", "Heun", "DDIM"])
                        strength = gr.Slider(label="Strength", value=0.7, minimum=0, maximum=1, step=0.001)
                        seed = gr.Slider(label="Seed", value=-1, minimum=-1, maximum=1000000000, step=1)
                        huggingface_api_key = gr.Textbox(label="Hugging Face API Key (required for API calls)", placeholder="Enter your Hugging Face API Key here", type="password", elem_id="api-key")
                        use_dev = gr.Checkbox(label="Use Dev API", value=False, elem_id="use-dev-checkbox")
                        enhance_prompt_style =  gr.Textbox(label="Enhance Prompt Style", placeholder="Enter style for the prompt enhancer here", elem_id="enhance-prompt-style")
                        enhance_prompt_option = gr.Checkbox(label="Enhance Prompt", value=False, elem_id="enhance-prompt-checkbox")
                        use_mistral_nemo = gr.Checkbox(label="Use Mistral Nemo", value=False, elem_id="use-mistral-checkbox")
                        nemo_prompt_style =  gr.Textbox(label="Nemo Enhance Prompt Style", placeholder="Enter style for the prompt enhancer here", elem_id="nemo-enhance-prompt-style")
                        
        with gr.Row():
            text_button = gr.Button("Run", variant='primary', elem_id="gen-button")
        with gr.Row():
            image_output = gr.Image(type="pil", label="Image Output", elem_id="gallery")
        with gr.Row():
            seed_output = gr.Textbox(label="Seed Used", elem_id="seed-output")
            final_prompt_output = gr.Textbox(label="Final Prompt", elem_id="final-prompt-output")
        
        inputs=[text_prompt, negative_prompt, steps, cfg, method, seed, strength, huggingface_api_key, use_dev, enhance_prompt_style,enhance_prompt_option, enhance_prompt_style, use_mistral_nemo]
        
        # Adjust the click function to include the API key, use_dev, and enhance_prompt_option as inputs
        text_button.click(query, inputs=inputs, outputs=[image_output, seed_output, final_prompt_output])

app.launch(show_api=True, share=False, show_error=True)