Add documentation for external sensitive media detection API

- Create API documentation explaining request/response format
- Add JSDoc comments to AiService methods
- Provide example implementation using nsfwjs

Co-authored-by: syuilo <4439005+syuilo@users.noreply.github.com>

docs/sensitive-media-detection-api.md

@@ -0,0 +1,117 @@
+# Sensitive Media Detection External Service API
+
+This document describes the API contract for an external sensitive media detection service that can be used with Misskey.
+
+## Endpoint
+
+The external service should expose a single HTTP POST endpoint at the URL configured in the Misskey admin settings (`sensitiveMediaDetectionProxyUrl`).
+
+## Request
+
+### Method
+`POST`
+
+### Headers
+- `Content-Type: application/json`
+
+### Body
+JSON object with the following structure:
+```json
+{
+  "image": "base64-encoded-image-data"
+}
+```
+
+Where `image` is the base64-encoded binary data of the image file to analyze.
+
+## Response
+
+### Success (HTTP 200)
+
+JSON object with the following structure:
+```json
+{
+  "predictions": [
+    {
+      "className": "Neutral",
+      "probability": 0.95
+    },
+    {
+      "className": "Drawing",
+      "probability": 0.03
+    },
+    {
+      "className": "Sexy",
+      "probability": 0.01
+    },
+    {
+      "className": "Porn",
+      "probability": 0.005
+    },
+    {
+      "className": "Hentai",
+      "probability": 0.005
+    }
+  ]
+}
+```
+
+The `predictions` array should contain objects with:
+- `className`: One of `"Neutral"`, `"Drawing"`, `"Sexy"`, `"Porn"`, or `"Hentai"`
+- `probability`: A number between 0 and 1 indicating the confidence level
+
+The classification names and behavior should match the [nsfwjs](https://github.com/infinitered/nsfwjs) library.
+
+### Error
+
+On error, the external service can return any HTTP error status code. Misskey will treat the request as failed and will not mark the file as sensitive (allowing the upload to proceed).
+
+## Timeout
+
+Requests will timeout after 10 seconds. The external service should respond within this time limit.
+
+## Example Implementation
+
+A simple example using nsfwjs:
+
+```javascript
+const express = require('express');
+const nsfw = require('nsfwjs');
+const tf = require('@tensorflow/tfjs-node');
+
+const app = express();
+app.use(express.json({ limit: '50mb' }));
+
+let model;
+
+async function loadModel() {
+  model = await nsfw.load();
+}
+
+app.post('/', async (req, res) => {
+  try {
+    const { image } = req.body;
+    const buffer = Buffer.from(image, 'base64');
+    const tfImage = await tf.node.decodeImage(buffer, 3);
+    const predictions = await model.classify(tfImage);
+    tfImage.dispose();
+    
+    res.json({ predictions });
+  } catch (error) {
+    res.status(500).json({ error: error.message });
+  }
+});
+
+loadModel().then(() => {
+  app.listen(3000, () => {
+    console.log('Sensitive media detection service running on port 3000');
+  });
+});
+```
+
+## Benefits of External Service
+
+- **Reduced Memory Usage**: nsfwjs and TensorFlow can be memory-intensive. Running them separately reduces Misskey's memory footprint.
+- **Scalability**: The detection service can be scaled independently.
+- **Alternative Implementations**: Operators can use different detection models or services (e.g., cloud-based APIs, more efficient models).
+- **Isolation**: Issues with the detection service don't affect Misskey's core functionality.

packages/backend/src/core/AiService.ts

@@ -35,6 +35,16 @@ export class AiService {
 	) {
 	}
 
+	/**
+	 * Detect sensitive content in an image.
+	 * 
+	 * If an external proxy URL is configured (sensitiveMediaDetectionProxyUrl),
+	 * this method will use the external service. Otherwise, it will use the
+	 * built-in nsfwjs model.
+	 * 
+	 * @param source - Path to the image file or a Buffer containing the image data
+	 * @returns Array of predictions or null if detection fails
+	 */
 	@bindThis
 	public async detectSensitive(source: string | Buffer): Promise<nsfw.PredictionType[] | null> {
 		// If external service is configured, use it
@@ -78,6 +88,16 @@ export class AiService {
 		}
 	}
 
+	/**
+	 * Detect sensitive content using an external proxy service.
+	 * 
+	 * Sends the image as base64-encoded data to the external service
+	 * and expects a response in the same format as nsfwjs.
+	 * 
+	 * @param source - Path to the image file or a Buffer containing the image data
+	 * @returns Array of predictions or null if the request fails
+	 * @see docs/sensitive-media-detection-api.md for API contract details
+	 */
 	@bindThis
 	private async detectSensitiveWithProxy(source: string | Buffer): Promise<nsfw.PredictionType[] | null> {
 		try {