@ekipnico/image-mod

AI-powered NSFW and content moderation for images.

Installation

npm install @ekipnico/image-mod

Quick Start

import { createImageModMesh, ImageModerator } from '@ekipnico/image-mod';

const mesh = createImageModMesh();
const moderator = mesh.resolve(ImageModerator);

const result = await moderator.check(imageBuffer);
// { safe: true, flagged: [], scores: { adult: 0.1, violence: 0.0, ... } }

Model Configuration

Models can be specified in two ways:

1. String ID (Built-in Providers)

Use string IDs for OpenAI, Anthropic, and Google models:

// Via environment variable (recommended for defaults)
process.env.AI_MESH_DEFAULT_MODEL = 'gpt-4o';

// Or per-request
const result = await moderator.check(imageBuffer, { model: 'gemini-flash' });

Built-in models: gpt-4o, gpt-4o-mini, gemini-flash, gemini-2.0-flash, claude-sonnet, etc.

2. LanguageModel (Any Provider)

Pass any Vercel AI SDK model directly for providers like Groq, Mistral, DeepSeek, etc:

import { createGroq } from '@ai-sdk/groq';

const groq = createGroq({ apiKey: process.env.GROQ_API_KEY });
const result = await moderator.check(imageBuffer, {
    model: groq('llama-3.3-70b-versatile'),
});

import { createMistral } from '@ai-sdk/mistral';

const mistral = createMistral({ apiKey: process.env.MISTRAL_API_KEY });
const result = await moderator.check(imageBuffer, {
    model: mistral('pixtral-large-latest'),
});

Environment Variables

| Variable | Description | |----------|-------------| | AI_MESH_DEFAULT_MODEL | Default model when not specified (default: gpt-4o) | | OPENAI_API_KEY | Required for OpenAI models | | ANTHROPIC_API_KEY | Required for Anthropic models | | GOOGLE_API_KEY | Required for Google/Gemini models |

API

moderator.check(input, config?)

Checks an image for moderation issues.

Parameters:

• input - ImageInput (Buffer, URL, file path, or ImageInput object)
• config.model - ModelSpec (string ID or LanguageModel)
• config.adult - Threshold for adult content (0-1, default: 1)
• config.violence - Threshold for violence (0-1, default: 1)
• config.racy - Threshold for racy content (0-1, default: 1)
• config.medical - Threshold for medical content (0-1, default: 0.5)
• config.spoof - Threshold for manipulated content (0-1, default: 0.5)

Threshold behavior:

• 0 = Skip this category
• 0.01-1 = Flag if score >= threshold (higher = stricter)

Returns:

{
    safe: boolean;           // true if no categories flagged
    flagged: string[];       // Categories that exceeded thresholds
    scores: ModerationScores // Raw scores (0-1) for each category
}

Integrating into Existing Mesh

import { Mesh } from 'mesh-ioc';
import { registerImageModServices, ImageModerator } from '@ekipnico/image-mod';

const mesh = new Mesh('MyApp');
// ... register your services ...
registerImageModServices(mesh);

const moderator = mesh.resolve(ImageModerator);