app.py

import datetime
import gradio as gr
import requests
import random
import io
import zipfile
from PIL import Image
import os
import numpy as np
import json
import boto3


# Create an S3 client
s3 = boto3.client('s3')

def save_to_s3(image_data, payload, file_name):
    # Define the bucket and the path
    bucket_name = 'dataset-novelai'
    folder_name = datetime.datetime.now().strftime("%Y-%m-%d")
    image_key = f'gradio/{folder_name}/{file_name}.webp'
    payload_key = f'gradio/{folder_name}/{file_name}.json'

    # Save the image
    image_data.seek(0)  # Go to the start of the BytesIO object
    s3.upload_fileobj(image_data, bucket_name, image_key, ExtraArgs={'ContentType': 'image/webp'})

    # Save the payload
    payload_data = io.BytesIO(payload.encode('utf-8'))
    s3.upload_fileobj(payload_data, bucket_name, payload_key, ExtraArgs={'ContentType': 'application/json'})


# Function to handle the NovelAI API request
def generate_novelai_image(input_text, quality_tags, seed, negative_prompt, scale, ratio, sampler):
    jwt_token = os.environ.get('NAI_API_KEY')
    if ratio == "Landscape (1216x832)":
        width = 1216
        height = 832
    elif ratio == "Square (1024x1024)":
        width = 1024
        height = 1024
    elif ratio == "Portrait (832x1216)":
        width = 832
        height = 1216
        
    # Check if quality tags are provided and append to input
    final_input = input_text
    if quality_tags:
        final_input += ", " + quality_tags

    # Assign a random seed if seed is -1
    if seed == -1:
        seed = random.randint(0, 2**32 - 1)

    # Define the API URL
    url = "https://api.novelai.net/ai/generate-image"

    # Set the headers
    headers = {
        "Authorization": f"Bearer {jwt_token}",
        "Content-Type": "application/json",
        "Origin": "https://novelai.net",
        "Referer": "https://novelai.net/"
    }

    # Define the payload
    payload = {
        "action": "generate",
        "input": final_input,
        "model": "nai-diffusion-3",
        "parameters": {
            "width": width,
            "height": height,
            "scale": scale,
            "sampler": sampler,
            "steps": 28,
            "n_samples": 1,
            "ucPreset": 0,
            "add_original_image": False,
            "cfg_rescale": 0,
            "controlnet_strength": 1,
            "dynamic_thresholding": False,
            "legacy": False,
            "negative_prompt": negative_prompt,
            "noise_schedule": "native",
            "qualityToggle": True,
            "seed": seed,
            "sm": False,
            "sm_dyn": False,
            "ucPreset": 0,
            "uncond_scale": 1,
        }
    }

    # Send the POST request
    response = requests.post(url, json=payload, headers=headers)

    # Process the response
    if response.headers.get('Content-Type') == 'application/x-zip-compressed':
        zipfile_in_memory = io.BytesIO(response.content)
        with zipfile.ZipFile(zipfile_in_memory, 'r') as zip_ref:
            file_names = zip_ref.namelist()
            if file_names:
                with zip_ref.open(file_names[0]) as file:
                    image = Image.open(file)

                    # Prepare to save the image to S3
                    buffered = io.BytesIO()
                    image.save(buffered, format="WEBP", quality=98)
                    file_name = str(int(datetime.datetime.now().timestamp()))
                    save_to_s3(buffered, json.dumps(payload, indent=4), file_name)

                    return np.array(image), json.dumps(payload, indent=4)

            else:
                return "No images found in the zip file.", json.dumps(payload, indent=4)
    else:
        return "The response is not a zip file.", json.dumps(payload, indent=4)

# Create Gradio interface
iface = gr.Interface(
    fn=generate_novelai_image,
    inputs=[
        gr.Textbox(label="Input Text"),
        gr.Textbox(label="Quality Tags", value="best quality, amazing quality, very aesthetic, absurdres"),
        gr.Slider(minimum=-1, maximum=2**32 - 1, step=1, value=-1, label="Seed"),
        gr.Textbox(label="Negative Prompt", value="nsfw, lowres, {bad}, error, fewer, extra, missing, worst quality, jpeg artifacts, bad quality, watermark, unfinished, displeasing, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]"),
        gr.Slider(minimum=1, maximum=20, step=1, value=5, label="Scale"),
        gr.Radio(choices=["Landscape (1216x832)", "Square (1024x1024)", "Portrait (832x1216)"], value="Portrait (832x1216)"),
        gr.Dropdown(
            choices=[
                "k_euler", "k_euler_ancestral", "k_dpmpp_2s_ancestral", 
                "k_dpmpp_2m", "k_dpmpp_sde", "ddim_v3"
            ],
            value="k_euler",
            label="Sampler"
        )
    ],
    outputs=[
        "image",
        gr.Textbox(label="Submitted Payload")
    ]
)

try:
    iface.queue(concurrency_count=3).launch(share=True)
except RuntimeError: # use in HF spaces
    iface.queue(concurrency_count=3).launch()