app.py

import gradio as gr
import requests
import base64
import os
import io
from PIL import Image

# --- OpenAI API Call Logic ---

def call_openai_image_api(prompt: str, api_key: str, input_image: Image.Image | None = None):
    """
    Calls the appropriate OpenAI Image API (generation or edit)
    based on whether an input image is provided.

    Args:
        prompt: The text prompt for image generation or editing.
        api_key: The OpenAI API key.
        input_image: A PIL Image object for editing, or None for generation.

    Returns:
        A tuple containing:
        - original_image (PIL.Image or None): The original image if editing, else None.
        - result_image (PIL.Image or None): The generated/edited image, or None on error.
        - status_message (str): A message indicating success or error details.
    """
    if not api_key:
        return None, None, "Error: OpenAI API Key is missing."
    if not prompt:
        return None, None, "Error: Prompt cannot be empty."

    headers = {"Authorization": f"Bearer {api_key}"}
    # Hypothetical model name from the original code. Replace with "dall-e-2" or "dall-e-3" if needed.
    model = "gpt-image-1" # Using the model specified in the original code
    size = "1024x1024"
    response = None # Initialize response variable

    try:
        if input_image:
            # --- Image Editing ---
            if not isinstance(input_image, Image.Image):
                 return None, None, "Error: Invalid image provided for editing."

            # Convert PIL Image to bytes for the API request
            byte_stream = io.BytesIO()
            input_image.save(byte_stream, format="PNG") # Save PIL image to bytes buffer [[1]]
            byte_stream.seek(0) # Rewind buffer to the beginning

            files = {
                "image": ("input_image.png", byte_stream, "image/png"),
            }
            # CORRECTED data dictionary: removed 'response_format'
            data = {
                "prompt": prompt,
                "model": model,
                "size": size,
                # "response_format": "b64_json", # <-- THIS LINE IS REMOVED
            }
            api_url = "https://api.openai.com/v1/images/edits"
            print("Calling OpenAI Image Edit API...") # Debug print
            response = requests.post(api_url, headers=headers, files=files, data=data)

        else:
            # --- Image Generation ---
            # (This part remains the same as it uses response_format correctly via json payload)
            headers["Content-Type"] = "application/json"
            payload = {
                "prompt": prompt,
                "model": model,
                "response_format": "b64_json", # Keep this for generation
                "size": size,
                "n": 1, # Generate one image
            }
            api_url = "https://api.openai.com/v1/images/generations"
            print("Calling OpenAI Image Generation API...") # Debug print
            response = requests.post(api_url, headers=headers, json=payload)

        print(f"API Response Status Code: {response.status_code}") # Debug print
        response.raise_for_status() # Raise HTTPError for bad responses (4xx or 5xx)

        # Process successful response
        response_data = response.json()
        # Ensure the expected data structure is present
        if not response_data.get("data") or not isinstance(response_data["data"], list) or len(response_data["data"]) == 0:
            return input_image, None, f"Error: Unexpected API response format - 'data' array missing or empty: {response_data}"
        if not response_data["data"][0].get("b64_json"):
             return input_image, None, f"Error: Unexpected API response format - 'b64_json' key missing: {response_data}"

        img_b64 = response_data["data"][0]["b64_json"]
        img_bytes = base64.b64decode(img_b64) # Decode base64 string [[1]]
        result_image = Image.open(io.BytesIO(img_bytes)) # Convert bytes to PIL Image [[1]]
        print("Image processed successfully.") # Debug print
        return input_image, result_image, "Success!"

    except requests.exceptions.RequestException as e:
        error_message = f"API Request Error: {e}"
        # Check if response exists before trying to access its attributes/methods
        if response is not None:
            try:
                # Attempt to get more specific error from OpenAI response
                error_detail = response.json()
                error_message += f"\nAPI Error Details: {error_detail}"
            except requests.exceptions.JSONDecodeError:
                # Fallback if response is not JSON
                error_message += f"\nRaw Response Text: {response.text}"
            except Exception as json_e:
                 error_message += f"\nError parsing JSON response: {json_e}\nRaw Response Text: {response.text}"
        print(error_message) # Debug print
        return input_image, None, error_message
    except Exception as e:
        error_message = f"An unexpected error occurred: {e}"
        print(error_message) # Debug print
        return input_image, None, error_message


# --- Gradio Interface Setup ---

# Check for API key in environment variables
api_key_env = os.environ.get("OPENAI_API_KEY")
api_key_present_info = "OpenAI API Key found in environment variables." if api_key_env else "OpenAI API Key not found in environment variables. Please enter it below."

def process_image_request(prompt_input, api_key_input, uploaded_image):
    """
    Wrapper function for Gradio interface.
    Determines the API key to use (input field first, then environment variable).
    Calls the main API function.
    """
    # Prioritize the API key entered in the input field
    final_api_key = api_key_input if api_key_input else api_key_env

    # Call the actual API logic
    original_img, result_img, status = call_openai_image_api(prompt_input, final_api_key, uploaded_image)

    # Return values for the Gradio output components
    # If generating (original_img is None), return None for the original image display
    # If editing, return the uploaded image (original_img) for the original image display
    return original_img, result_img, status


# Build the Gradio interface using Blocks for more layout control [[7]]
with gr.Blocks() as demo:
    gr.Markdown("# OpenAI GPT-Image-1 Text-to-Image Demo") # App title
    gr.Markdown("Enter a prompt to generate an image, or upload an image and enter a prompt to edit it.")

    with gr.Row():
        with gr.Column(scale=1):
            prompt_input = gr.Textbox(label="Image Description (Prompt)", lines=3, placeholder="e.g., A futuristic cityscape at sunset") # Text input for prompt
            gr.Markdown(f"*{api_key_present_info}*")
            api_key_input = gr.Textbox(label="OpenAI API Key", type="password", placeholder="Enter your key if not set in environment") # Password input for API key
            uploaded_image_input = gr.Image(type="pil", label="Upload Image to Edit (Optional)") # Image upload [[4]]
            submit_button = gr.Button("Generate / Edit Image")
        with gr.Column(scale=2):
            status_output = gr.Textbox(label="Status", interactive=False)
            with gr.Row():
                 original_image_output = gr.Image(label="Original Image", interactive=False)
                 result_image_output = gr.Image(label="Generated / Edited Image", interactive=False) # Display output image

    # Connect the button click event to the processing function
    submit_button.click(
        fn=process_image_request,
        inputs=[prompt_input, api_key_input, uploaded_image_input],
        outputs=[original_image_output, result_image_output, status_output]
    )

# Launch the Gradio app [[2]]
if __name__ == "__main__":
    demo.launch()