Content moderation with Hive

Add content moderation with Hive into your video platform in a few clicks with AVflow

You need content moderation for your videos to determine if your videos are safe or NSFW (not safe for work). With Hive + AVflow, you can now add content mod into your application in a few minutes!

Read below to manually set this up or clone a similar flow (that uses Hive and Mux) to get started even quicker.

Summary: a video gets uploaded to an S3 bucket which triggers the flow; this video is sent to Hive for analysis. Hive returns a JSON result identifying parts of the video which gets flagged for certain categories (such as nudity, violence, etc) with a probability score. Then AVflow runs a script where you can analyze the JSON results for scores that exceed a threshold and determines the video to be "safe" or "unsafe". This result can then be saved to a Table or sent to the client application via a webhook where the client app can decide, such as, to not to make a streaming URL available to the public.

Here's how to set this up:

1. Add "Hive" as a step to the Flow

2. Setup the step options

Video Source: The source video which will need to be checked before uploading

Classes Filter: The array of tags/labels (aka classes) that need to be checked (["nsfw", "sexual"] ), if not provided this step will return all the tags found/detected in the video. If provided, the result will return the specified tag/labels with the associated scores.

Here's an example:

["nsfw", "general_nsfw", "yes_realistic_nsfw", "yes_male_nudity", "yes_female_nudity"]

Note: Need to use double quote for each tag string.

The Hive API Key: The Hive API key associated with the account/project.

3. Add Transfer to Storage step to save the JSON result file to S3 so later steps can process it.

4. Function script: Add a function to read Hive's output JSON to see if there's a label indicating that content is safe or unsafe. You can use our script if you would like or clone it from here:

const https = require('https');
const fs = require('fs');
const request = require('request');

async function downloadFromS3(srcUrl, dstFile) {
    var downloaded = 0;

    return new Promise((resolve, reject) => {
        const file = fs.createWriteStream(dstFile);
        const request = https.get(srcUrl, (response) => {
            response.pipe(file);
            
            response.on('data',(chunk) => {
                downloaded += chunk.length;
            })
            file.on('finish', (e) => {
                console.log('Saved to: ' + dstFile);
                console.log('File length: ' + downloaded);
                file.end();
                resolve({
                    fileName: dstFile,
                    length: downloaded
                });
            });   
        });
    });    
}

async function checkingThreshold(theHiveResult, threshold) {
    if(!threshold) return false;

    for (var i = 0; i < theHiveResult.length; i++){
        for (var j = 0; j < theHiveResult[i].classes.length; j++) {
            if(theHiveResult[i].classes[j].score < parseFloat(threshold)) {
                console.log("1111");
                return false;
            }
        }
    };
    return true;
}

async function main (service, context) {
    try {
        const threshold = service.threshold;
    
        const local_thehive_file = './thehive.json'
        await downloadFromS3(service.theHiveJSON, local_thehive_file);
        
        const objectResultTheHive = JSON.parse(fs.readFileSync(local_thehive_file, 'utf8'));
        
        
        const checking = await checkingThreshold(objectResultTheHive.output, threshold)

        return {result : checking ? "safe" : "unsafe"}
    } catch(e) {
        console.log("ERROR SCRIPTING", e);
        return e;
    }
}

module.exports = main;

Input parameters:

[  {    dataTypeId: 4,    supportedDataTypeIds: [      '4',      '16',      '14',      '12',      '15',      '1',      '2',      '3',      '6',      '7'    ],    require: true,    key: 'theHiveJSON',    dataType: 'string'  },  {    dataTypeId: 4,    supportedDataTypeIds: [      '4',      '16',      '14',      '12',      '15',      '1',      '2',      '3',      '6',      '7'    ],    require: true,    key: 'threshold',    dataType: 'string'  }]

Output Parameters:

[  {    dataTypeId: 4,    key: 'result',    dataType: 'string'  }]

5. Webhook: Then you can add a webhook back to your client app to decide if you want to make a video private. Here is an example:

6. Tada! Turn the Flow on and check the logs to see the results.

‍