Practical overview
When it fits
Use Nsfwjs when an application needs an image classification step inside a JavaScript client or runtime and the developer can handle the returned probabilities in the surrounding workflow.
Who it is for
- Developers building browser-based image upload workflows
- Teams adding local content screening to image-sharing applications
- JavaScript developers working with browser, extension or Node.js environments
Documented uses
- Classify an image into documented content categories
- Screen images before they are submitted or published
- Add a local first-pass check to user-generated image workflows
- Load bundled or self-hosted TensorFlow.js models for image classification
Requirements & limits
The project is publicly available as an MIT-licensed repository. The supplied material does not document pricing or an account requirement for using the library.
- Client-side execution can keep the classification step in the browser rather than relying on a hosted moderation service.
- The library supports MobileNetV2, MobileNetV2Mid and InceptionV3 model options.
- Developers can use bundled models or host model files at their own URL.
- TensorFlow.js backend guidance includes WASM, WebGPU and Node.js options.
- Model dimensions and graph-model handling vary by selected model.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Add a browser image-classification check
Load an Nsfwjs model and classify an image element in a JavaScript application.
Before you start
- A JavaScript or TypeScript application
- The nsfwjs package and TensorFlow.js environment
- An image element available to the application
Steps
Import nsfwjs and obtain the image element.
Load the default model with nsfwjs.load(), or select a documented bundled model.
Call model.classify(img).
Inspect the returned predictions for the five documented categories.
Expected outcome
The application receives category probabilities for the supplied image.
Guide limitations
- The project states that classification is not perfect.
- The image must match the requirements of the selected model; documented bundled models include MobileNetV2 at 224×224 and InceptionV3 at 299×299.
- This recipe describes classification, not a complete publication or enforcement workflow.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 2 October 2026
- Original website title
- infinitered/nsfwjs · Official homepage title · retained in its source language