README.md

Echikana 🫣

A Node.js library for NSFW (not safe for work) classification of images using Falconsai/nsfw_image_detection.

import fs from "node:fs";
import { EchikanaInferencer } from "echikana";

const model = fs.readFileSync("path/to/model.onnx");
const inferencer = new EchikanaInferencer(model);

await inferencer.initialize();

const image = fs.readFileSync("path/to/image.jpg");
const result: number = await inferencer.inference(image);

console.log(`The image is ${Math.round(result * 100)}% NSFW!`);

API

First, instantiate the EchikanaInferencer class by passing the model data (ArrayBufferLike) as an argument to the constructor function.

import fs from "node:fs";
import { EchikanaInferencer } from "echikana";

const model = fs.readFileSync("path/to/model.onnx");
// const model = await fetch("https://www.example.com/path/to/model.onnx").then((response) => response.arrayBuffer());

const inferencer = new EchikanaInferencer(model);

Next, call the initialize method to initialize the model. The model will not be loaded until this method is called.

await inferencer.initialize();

Now, let's perform inference! The inference method accepts an image as an ArrayBufferLike and returns a fraction from 0 to 1 indicating the NSFW probability of the image.

const image = fs.readFileSync("path/to/image.jpg");
const result: number = await inferencer.inference(image);

console.log(`The image is ${Math.round(result * 100)}% NSFW!`);

Note

During inference, the image is transformed using sharp for input to the model. The image formats that can be converted are limited to those supported by sharp.

If you need to discard a loaded model, use the dispose method.

To reload the model, call the initialize method again. The previously loaded model will be disposed of and newly initialized.

About the inference model

This library uses onnxruntime-node as the runtime for model execution. Therefore, to use the “Falconsai/nsfw_image_detection” model published on Hugging Face, you must first convert it to ONNX format.

Warning

No inference model is included in this package or repository.

Please use Optimum, a conversion tool by Hugging Face.

optimum-cli export onnx --model Falconsai/nsfw_image_detection /path/to/model-dir

This repository includes a Dockerfile and a compose.yaml file, which are useful for converting the model to ONNX format. To use them, run the following commands.

git clone https://github.com/okayurisotto/echikana.git echikana
cd echikana
docker compose run --rm --build optimum

The converted model file will be created in the model directory.