The NSFW Question

With a worker joining the horde recently who was accidentally censoring NSFW generations without meaning to, it pressed me to implement the solution I was considering for the NSFW question.

To the question is: Does the Horde allow NSFW?

The answer must be: Yes, but voluntarily!

As the horde is effectively just a dumb pipe between workers and clients, barely one step above pure p2p, it is impossible for me to figure out which requests or generations are NSFW. In fact, way more powerful companies than myself have tried and failed to do the same, often completely destroying their service in the process.

As I do not have the resources or the capability to filter NSFW myself (I AM a solodev after all), like the generations themselves, I have to offload this decision, to the community itself.

So the recent update to the bridge adds two new fields to the client request APIs and 1 new field to the worker API.

The worker's bridge now adds a new "nsfw" variable, which is sent to the Horde when the worker checks-in. It defaults to True which allows the worker to serve all requests like before. However if this is set to False, then this worker will only pick-up requests which are marked as SFW and will skip all NSFW requests. So if you, as the owner of the worker, don't feel comfortable serving NSFW request, just flip the switch and you're safe. And if you want to ensure that nobody sends a NSFW requests marked as SFW, you can enable the NSFW filter on your end, which will return a black image on NSFW generations/

Likewsie, the clients can now specify whether what they're requesting is supposed to be NSFW on purpose. This is the new "nsfw" boolean key. If set to False (default), Then it works as right now. The request will go to the first available worker. However if the request has the potential to be NSFW (or is explicitly prompted to be so), and the worker has enabled NSFW filters, you might get some censored images back.

To avoid this, set NSFW to True in your requests, and thus only workers which allow NSFW will pick up your request and you will ensure no censorship. If you anyway see censored images this way, contact me!

Furthermore, if you're sending a potentially NSFW prompt marked as SFW, but you want to ensure you do not see NSFW, you can also set the "censor_nsfw" to True (default). This will tell all workers (including NSFW ones), that they should censor any NSFW generations they did with your prompt. This way you can send a risky prompt like "saucy hot woman" to your generations, and ensure you don't get any parts you don't want to see ;)

The "censor_nsfw" is ignored if you mark your generation as "nsfw".

I believe this approach can satisfy all parties now. Workers can ensure they generate only what they're comfortable with, and clients can mark their requests accordingly, with an extra safety belt for "accidents".

And of course, if someone malicious joins and tries to bypass these measures on purpose, please do inform me!