api/index.py

import os
import base64
import numpy as np
from PIL import Image
import io
import requests

import replicate
from flask import Flask, request
import gradio as gr
import openai
from openai import OpenAI

from dotenv import load_dotenv, find_dotenv

# Locate the .env file
dotenv_path = find_dotenv()

load_dotenv(dotenv_path)

OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
REPLICATE_API_TOKEN = os.getenv('REPLICATE_API_TOKEN')

client = OpenAI()

def call_openai(pil_image):
    # Save the PIL image to a bytes buffer
    
    buffered = io.BytesIO()
    pil_image.save(buffered, format="JPEG")
    
    # Encode the image to base64
    image_data = base64.b64encode(buffered.getvalue()).decode('utf-8')
    
    try:
        response = client.chat.completions.create(
            model="gpt-4o",
            messages=[
                {
                    "role": "user",
                    "content": [
                        {"type": "text", "text": "You are a product designer. I've attached a moodboard here. In one sentence, what do all of these elements have in common? Answer from a design language perspective, if you were telling another designer to create something similar, including any repeating colors and materials and shapes and textures"},
                        {
                            "type": "image_url",
                            "image_url": {
                                "url": "data:image/jpeg;base64," + image_data,
                            },
                        },
                    ],
                }
            ],
            max_tokens=300,
        )
        
        return response.choices[0].message.content

    except openai.BadRequestError as e:
        print(e)
        print("e type")
        print(type(e))
        raise gr.Error(f"Please retry with a different moodboard file")
    except Exception as e:
        raise gr.Error("Unknown Error")

def image_classifier(moodboard, starter_image, image_strength, prompt):
    raise gr.Error(header)
    
    if moodboard is not None and starter_image is not None:

        # Convert the numpy array to a PIL image
        pil_image = Image.fromarray(moodboard.astype('uint8'))
        starter_image_pil = Image.fromarray(starter_image.astype('uint8'))

        # Resize the starter image if either dimension is larger than 768 pixels
        if starter_image_pil.size[0] > 768 or starter_image_pil.size[1] > 768:
            # Calculate the new size while maintaining the aspect ratio
            if starter_image_pil.size[0] > starter_image_pil.size[1]:
                # Width is larger than height
                new_width = 768
                new_height = int((768 / starter_image_pil.size[0]) * starter_image_pil.size[1])
            else:
                # Height is larger than width
                new_height = 768
                new_width = int((768 / starter_image_pil.size[1]) * starter_image_pil.size[0])
            
            # Resize the image
            starter_image_pil = starter_image_pil.resize((new_width, new_height), Image.LANCZOS)
        
        #openai_response = call_openai(pil_image)
        #openai_response = openai_response.replace('moodboard', '')
        #openai_response = openai_response.replace('share', '')
        #openai_response = openai_response.replace('unified', '')

        # Save the starter image to a bytes buffer
        buffered = io.BytesIO()
        starter_image_pil.save(buffered, format="JPEG")
        
        # Encode the starter image to base64
        starter_image_base64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
    else:
        raise gr.Error(f"Please upload a moodboard to control image generation style")
        
    # Call Stable Diffusion API with the response from OpenAI
    input = {
        "width": 768,
        "height": 768,
        "prompt": "high quality render of " + prompt + ", " + openai_response[12:],
        "negative_prompt": "worst quality, low quality, illustration, 2d, painting, cartoons, sketch",
        "refine": "expert_ensemble_refiner",
        "image": "data:image/jpeg;base64," + starter_image_base64, 
        "apply_watermark": False,
        "num_inference_steps": 25,
        "prompt_strength": 1-image_strength,
        "num_outputs": 3
    }
    
    output = replicate.run(
        "stability-ai/sdxl:7762fd07cf82c948538e41f63f77d685e02b063e37e496e96eefd46c929f9bdc",
        input=input
    )
    
    images = []
    for i in range(min(len(output), 3)):
        image_url = output[i]
        response = requests.get(image_url)
        images.append(Image.open(io.BytesIO(response.content)))
    
    # Add empty images if fewer than 3 were returned
    while len(images) < 3:
        images.append(Image.new('RGB', (768, 768), 'gray'))

    return images

header = "Set up APIs on HuggingFace or use free at https://app.idai.tools/ (https://app.idai.tools/interface/moodboard_controlled)"
demo = gr.Interface(fn=image_classifier, inputs=["image", "image", gr.Slider(0, 1, step=0.05, value=0.2, label="Image Strength"), "text"], outputs=["image", "image", "image"], title=header)
demo.launch(share=False)