app.py

import gradio as gr
from transformers import pipeline, WhisperProcessor, WhisperForConditionalGeneration
from diffusers import StableDiffusionPipeline
import torch
import numpy as np

# Step 1: Prompt-to-Prompt Generation using BART (or any LLM except GPT or DeepSeek)
prompt_generator = pipeline("text2text-generation", model="facebook/bart-large-cnn")

def generate_prompt(description: str) -> str:
    # Generate a detailed prompt based on a short description
    prompt = prompt_generator(f"Expand this description into a detailed prompt for an image: {description}", max_length=150)[0]['generated_text']
    return prompt

# Step 2: Prompt-to-Image Generation using Stable Diffusion v1.5 (with GPU/CPU Support)
device = "cuda" if torch.cuda.is_available() else "cpu"
stable_diffusion = StableDiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-1-base")
stable_diffusion.to(device)

def generate_image(prompt: str, creativity: float, include_background: bool):
    # Adjust creativity and background options in the prompt
    if creativity < 0.5:
        prompt += " with simpler details."
    else:
        prompt += " with highly detailed elements."

    if include_background:
        prompt += " with a vibrant and detailed background."
    
    # Generate image based on the prompt
    image = stable_diffusion(prompt).images[0]
    return image

# Step 3: Voice Input Integration using Whisper for Speech-to-Text
processor = WhisperProcessor.from_pretrained("openai/whisper-large")
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large")

def transcribe_audio(audio: np.ndarray, sampling_rate: int) -> str:
    # Directly process the numpy array audio input
    audio_input = processor(audio, sampling_rate=sampling_rate, return_tensors="pt").input_features
    predicted_ids = model.generate(audio_input)
    transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
    return transcription

# Step 4: Gradio Interface with Simple Controllers (Textbox, Slider, Checkbox, Audio)
def process_input(description: str, creativity: float, include_background: bool):
    # Generate a detailed prompt
    prompt = generate_prompt(description)
    
    # Generate image based on user inputs
    image = generate_image(prompt, creativity, include_background)
    
    return prompt, image

def process_audio_input(audio, sampling_rate):
    # Convert audio to text
    description = transcribe_audio(audio, sampling_rate)
    # Generate a prompt and image based on transcribed text
    prompt = generate_prompt(description)
    image = generate_image(prompt, creativity=0.7, include_background=True)
    return prompt, image

# Define Gradio interface components
text_input = gr.Textbox(label="Enter Description", placeholder="E.g., A magical treehouse in the sky")
creativity_slider = gr.Slider(minimum=0, maximum=1, step=0.1, label="Creativity (0 to 1)", value=0.7)
background_checkbox = gr.Checkbox(label="Include Background", value=True)

audio_input = gr.Audio(type="numpy", label="Speak your Description", source="microphone")

# Create Gradio interface for text input
interface = gr.Interface(
    fn=process_input,
    inputs=[
        text_input,
        creativity_slider,
        background_checkbox
    ],
    outputs=[
        gr.Textbox(label="Generated Prompt"),
        gr.Image(label="Generated Image")
    ],
    title="Magical Image Generator",
    description="Enter a short description to generate a magical image. Adjust creativity and background options.",
    theme="huggingface"
)

# Add audio input interface for voice interaction
interface_with_audio = gr.Interface(
    fn=process_audio_input,
    inputs=[audio_input],
    outputs=[gr.Textbox(label="Generated Prompt"), gr.Image(label="Generated Image")],
    title="Magical Image Generator with Voice Input",
    description="Speak a short description to generate a magical image!"
)

# Launch the interface with multiple tabs for text and voice input
gr.TabbedInterface([interface, interface_with_audio]).launch()