[RicoElectrico]

Can you consider giving a score instead of a binary decision? Or differentiate between, say, beach pics and true 18+ content.

Edit: The accuracy leaves a lot to be desired - you can paste any images from clothing stores of models in dresses or tank tops and it will flag it as NSFW.

The "AI" seems to replicate an Islamic fundamentalist - a woman in a burqa did pass as SFW ;)

Google offers a hosted SafeSearch version [1] which has a lot more nuance: https://cloud.google.com/vision/docs/reference/rpc/google.cl...

[lngzl]

It gives a score as well, in the API, I can for sure add that to what you see on the "Try out" page. I was planning on adding a button where you could see the raw data.

[swiley]

The whole thing is built to export moderation to an external service. It's a bit creepy IMO.

[lngzl]

I totally get that. That's why I'm planning on making the service (or at least parts of the service) open source. You can already get insights in the storage usage of the server through status.lngzl.nl (click scan.nsfw.rest). If you upload files, the storage size does not increase.

Also, one of the things I should make a little bit more clear is that it's best paired with data that will be made public anyways (for example, a social media platform) and shouldn't be run on private images (even though it's not storing anything it's probably not needed for those use cases).

[soared]

Corporate uses of nsfw filters typically want to filter out clothing store models and similar content. “Brand safe”