NSFW and adult content filter

We all know that receiving unsolicited NSFW content is never fun. Whether it’s a random gif of a naked person or an explicit photo, it’s always jarring (and often unwelcome). That’s why we are excited about our bot’s feature that can filter out NSFW content before it’s posted. The bot uses Artificial Intelligence models trained on 16 million photos and images to quickly identify and flag adult content. That way, you can rest assured that your Telegram channel will be safe for work (and for your eyeballs). So if you’re looking for a way to protect your channel from NSFW content, this is the droid you are looking for.

---

How the NSFW filter works?

The images, gifs and sticker filtering work seamlessly for you as a user. Whenever you enable the functionality in your group administrator panel bot starts looking for the media content in messages sent to your group. Whenever such content is found - it obtains the direct link to the media file and sends it to the images scanning microservice. It’s just the beginning of the story, as the first thing the NSFW check is doing - is verifying if the image is already known to the bot. It is to limit the usage of the relatively expensive processing power checks to the absolute and necessary minimum. It also saves your allowance if you use the premium. If an image, gif or sticker is new and not-known before, the microservice runs a scan using the AI model we have developed and trained on gigabytes of adult material. The whole process takes about a second, and it’s always a compromise between speed and accuracy, which we can proudly say is about 93%.

We check if the chance is higher than our requested threshold of “bot being sure”, which is currently set to 80% and issue the final verdict - should the user be punished and the message deleted or left as it is. After the check is complete ( or retried because of a networking issue ), we end up with the result in a clear form - identifying the chance of media being either adult material, drawings, NSFW drawings ( yes, they do exist! ) or a standard picture without any issues. The result is, of course, stored in the database together with a unique photo identifier. The next time the bot sees the same picture in any of the groups, it will not need to scan the content again, not touching your monthly allowance.

How about gif and video scanning?

We need to admit - that was a tricky one. Regular services divide gif into a few sections and check one frame from each. We know that spammers and trolls have discovered this and bypass it by modifying the gifs - adding normal pictures into the frames most commonly checked by those services. In our case - we are nice enough to check every single frame of the incoming gifs to be sure nothing unwanted has sneaked in. Then we do a lot of maths on the results trying to determine the overall rating for the whole gif ( not only the single frames ).

The bot made a mistake!

Whenever the bot detects the picture sent on the group ( or profile picture ) as inappropriate, it’s above us at the moment. Try to look at the content from the bot’s purely scientific perspective. Would you be able to point out what in this picture could identify it as a regular, not NSFW? Sometimes it could be a tattooed body in a swimsuit. In another case - a provocative pose or facial expression is synonymous with adult behaviour. Once the bot is in the wild - it makes the decisions independently. That’s why it’s called artificial intelligence, not the algorithmic approach.

If you are sure the bot made a mistake and want to avoid further kicks, limits, or even bans, you can contact us on the Official Support Chat and provide us with the penalty details. For legal reasons, we do not store any pictures scanned ( as we never know if malicious users send out some illegal content), so we can’t display them to you for assessment. Still, we can check the link to the file from the microservice logs and then manually override it in the database as safe for work.

What is the difference between pornography and indecency settings?

Pornography will detect most of the obvious pornographic materials posted into your group. Indecency covers things which can’t be considered a porn, but could be not really welcome in certain types of groups. As indecency bot will detect for example provocative poses, bikini, parts of the body etc. It’s essentially more strict porn filter for the bot.

What does NSFW filter scan?

You have a few options to choose from. The regular image, gifs and stickers scans are subject to the premium requirements and limits. Without any premium purchased, your group can benefit from the always free profile pictures scanning, which checks those users upload into their profiles for adult content. Other options may be enabled in your panel, but they will not work without a premium subscription.