|
|
|
|
|
|
by corysama
2789 days ago
|
|
|
If there’s too much manual work, you aren’t using enough layers! ;) Seems the smart approach for a... “focused” AI would be to manually label a bunch of nsfw bits and use that to build a censor bot that auto-detects and censors the of-interest regions. From there you can generate a large corpus of censored-uncensored pairs with which to train the decensor bot. |
|
|