app.py

import streamlit as st
# Set the page layout to 'wide'
st.set_page_config(layout="wide")
import requests
from PIL import Image
from io import BytesIO
# from IPython.display import display
import base64
import time



# helper decoder
def decode_base64_image(image_string):
    base64_image = base64.b64decode(image_string)
    buffer = BytesIO(base64_image)
    return Image.open(buffer)

# display PIL images as grid
def display_image(image=None,width=500,height=500):
    img = image.resize((width, height))
    return img

# API Gateway endpoint URL
api_url = 'https://a02q342s5b.execute-api.us-east-2.amazonaws.com/reinvent-demo-inf2-sm-20231114'

# Define the CSS to change the text input background color
input_field_style = """
<style>
    /* Customize the text input field background and text color */
    .stTextInput input {
        background-color: #fbd8bf; /* 'Rind' color */
        color: #232F3E; /* Dark text color */
    }
    /* You might also want to change the color for textarea if you're using it */
    .stTextArea textarea {
        background-color: #fbd8bf; /* 'Rind' color */
        color: #232F3E; /* Dark text color */
    }
</style>
"""

# Inject custom styles into the Streamlit app
st.markdown(input_field_style, unsafe_allow_html=True)


# Creating Tabs
tab1, tab2 = st.tabs(["Image Generation", "Text Generation"])

with tab1:
    # Create two columns for layout
    left_column, right_column = st.columns(2)
    # ===========
    with left_column:
        # Define Streamlit UI elements
        st.title('Stable Diffusion XL Image Generation with AWS Inferentia')

        prompt_one = st.text_area("Enter your prompt:", 
                            f"Raccoon astronaut in space, sci-fi, future, cold color palette, muted colors, detailed, 8k")

        # Number of inference steps
        num_inference_steps_one = st.slider("Number of Inference Steps", 
                                    min_value=1, 
                                    max_value=100, 
                                    value=30, 
                                    help="More steps might improve quality, with diminishing marginal returns. 30-50 seems best, but your mileage may vary.")

        # Create an expandable section for optional parameters
        with st.expander("Optional Parameters"):
            # Random seed input
            seed_one = st.number_input("Random seed", 
                                value=555, 
                                help="Set to the same value to generate the same image if other inputs are the same, change to generate a different image for same inputs.")

            # Negative prompt input
            negative_prompt_one = st.text_area("Enter your negative prompt:", 
                            "cartoon, graphic, text, painting, crayon, graphite, abstract glitch, blurry")

        





    if st.button('Generate Image'):
        with st.spinner(f'Generating Image with {num_inference_steps_one} iterations'):
            with right_column:
                start_time = time.time()
                # ===============
                # Example input data
                prompt_input_one = {
                    "prompt": prompt_one,
                    "parameters": {
                        "num_inference_steps": num_inference_steps_one,
                        "seed": seed_one,
                        "negative_prompt": negative_prompt_one
                    }
                }

                # Make API request
                response_one = requests.post(api_url, json=prompt_input_one)

                # Process and display the response
                if response_one.status_code == 200:
                    result_one = response_one.json()
                    # st.success(f"Prediction result: {result}")
                    image_one = display_image(decode_base64_image(result_one["generated_images"][0]))
                    st.image(image_one, 
                        caption=f"{prompt_one}") 
                    end_time = time.time()
                    total_time = round(end_time - start_time, 2)
                    st.text(f"Prompt: {prompt_one}")
                    st.text(f"Number of Iterations: {num_inference_steps_one}")
                    st.text(f"Random Seed: {seed_one}")
                    st.text(f'Total time taken: {total_time} seconds')
                    # Calculate and display the time per iteration in milliseconds
                    time_per_iteration_ms = (total_time / num_inference_steps_one)
                    st.text(f'Time per iteration: {time_per_iteration_ms:.2f} seconds')
                else:
                    st.error(f"Error: {response_one.text}")


with tab2:        
    # ===========
    # Define Streamlit UI elements
    st.title('Stable Diffusion XL Image Generation')



    prompt = st.text_area("Enter your prompt:", 
                        "Raccoons astronaut in space, sci-fi, future, cold color palette, muted colors, detailed, 8k")

    # Number of inference steps
    num_inference_steps = st.slider("Number of Inference Steps", 
                                min_value=1, 
                                max_value=100, 
                                value=40, 
                                help="More steps might improve quality, with diminishing marginal returns. 30-50 seems best, but your mileage may vary.")

    # Create an expandable section for optional parameters
    with st.expander("Optional Parameters"):
        # Random seed input
        seed = st.number_input("Random seed", 
                            value=42, 
                            help="Set to the same value to generate the same image if other inputs are the same, change to generate a different image for same inputs.")

        # Negative prompt input
        negative_prompt = st.text_area("Enter your negative prompt:", 
                        "anime, cartoon, graphic, text, painting, crayon, graphite, abstract glitch, blurry")

    





    if st.button('Generate Image:'):
        with st.spinner(f'Generating Image with {num_inference_steps} iterations'):
            # ===============
            # Example input data
            prompt_input = {
                "prompt": prompt,
                "parameters": {
                    "num_inference_steps": num_inference_steps,
                    "seed": seed,
                    "negative_prompt": negative_prompt
                }
            }

            # Make API request
            response = requests.post(api_url, json=prompt_input)

            # Process and display the response
            if response.status_code == 200:
                result = response.json()
                # st.success(f"Prediction result: {result}")
                image2 = display_image(decode_base64_image(result["generated_images"][0]))
                st.header("SDXL Base")
                st.image(image2, 
                    caption=f"SDXL Base, {num_inference_steps} iterations") 
            else:
                st.error(f"Error: {response.text}")