app.py

import gradio as gr
import tensorflow as tf
import numpy as np
from PIL import Image
from io import BytesIO
from tensorflow.keras.applications.inception_v3 import InceptionV3, preprocess_input
from tensorflow.keras.preprocessing.image import load_img, img_to_array
from transformers import TFAutoModelWithLMHead, AutoTokenizer
from stylegan2.tf_api import G_synthesis as StyleGAN2


# Load pre-trained image captioning model
tokenizer = AutoTokenizer.from_pretrained("t5-large")
model = TFAutoModelWithLMHead.from_pretrained("t5-large")

# Load pre-trained StyleGAN2 model
g = StyleGAN2()
g.load_weights('models/stylegan2-ffhq-config-f.pkl')

# Load pre-trained InceptionV3 model for image preprocessing
inception_v3 = InceptionV3(weights='imagenet')

# Define function to preprocess image for GAN
def preprocess_image(image):
    image = image.resize((256, 256))
    image_array = img_to_array(image)
    image_array = preprocess_input(image_array)
    image_array = np.expand_dims(image_array, axis=0)
    return image_array

# Define function to generate image from text using StyleGAN2
def generate_image(description):
    z = tf.random.normal([1, g.input_shape[1]])
    text = "generate image of a " + description
    input_ids = tokenizer.encode(text, return_tensors='tf')
    output = model.generate(input_ids=input_ids)
    caption = tokenizer.decode(output[0], skip_special_tokens=True)
    image = g(z, caption)
    image = (image.numpy()[0] * 255).astype(np.uint8)
    image = Image.fromarray(image, mode='RGB')
    return image

# Define function to generate text description of uploaded image using InceptionV3 and T5
def generate_description(image_file):
    image = Image.open(BytesIO(image_file.read()))
    image = preprocess_image(image)
    features = inception_v3.predict(image)
    features = tf.keras.backend.flatten(features)
    input_text = tokenizer.encode("generate a description of an image", return_tensors="tf")
    output = model.generate(input_ids=input_text, attention_mask=tf.ones(input_text.shape), max_length=50)
    caption = tokenizer.decode(output[0], skip_special_tokens=True)
    return caption

# Define functionto create the web application using Gradio
def image_generation(text_input, image_file):
    if image_file is not None:
        # Generate text description of uploaded image
        description = generate_description(image_file)
        # Generate image from text description
        generated_image = generate_image(description)
    else:
        # Generate image from user input text
        generated_image = generate_image(text_input)
    return generated_image

#Define Gradio interface
inputs = [gr.inputs.Textbox(label="Input text"),
          gr.inputs.Image(label="Upload an image (optional)")
         ]
outputs = gr.outputs.Image(label="Generated Image")

gr.Interface(
    fn=image_generation,
    inputs=inputs,
    outputs=outputs,
    title="Image Generation from Text",
    description="Generate high-quality images from text descriptions.",
    theme="default",
    layout="vertical",
    examples=[
["a red sports car on a mountain road"],
["a cute puppy"],
["an elegant woman with a hat and a scarf"],
["a scenic beach with palm trees and blue water"],
["a golden retriever sitting on a couch"],
["a delicious pizza with pepperoni and cheese"],
["a futuristic city with tall buildings and flying cars"],
["an adorable kitten playing with a ball of yarn"],
],
).launch(debug=True)