AI Dungeon Vore Potential

[LostCause]

We need a hero to create an AI.Dungeon like game for lewd and vore. Sadly i am not that hero ;-;

For lewd, NovelAI can be pretty effective. Especially if you can find a custom module or set of biases to support what you want. Unfortunately there has yet to be anything consistently effective in terms of vore, but I know of at least one person putting a HUGE amount of work into it for NovelAI (producing their own custom modules, lorebooks, phrase biases etc). Not only this but within the next half a year or possibly less Eleuther will likely release a more powerful AI that should aid in more niche usage such as for vore. Right now the AI can struggle with vore due to it requiring a lot of good quality examples within it's training data, which are of course lacking for vore. This problem should be alleviated somewhat by a more powerful AI in the future that doesn't require quite so many examples.

To re-clarify: NovelAI works great for lewd and has no filters or restrictions. You can do whatever you want with it and no one reads your stories or blocks outputs. It *can* currently work for vore too. However it requires a lot of fiddling, retrying and experimentation to get quality output for vore... and even then it won't be highly consistent at the moment. As for mixing vore AND sex... well in my experience the current AI model tends to confuse the two so personally I'd recommend trying to focus on one or the other at a time.

[MrXenomorph]

AID is kinda back on the redemtion path... and all... Latitude team (members of which were changed, quite a lot of them in fact) are slowly working their way to give up on ClosedAI entirely (from what I understand, they want to give up on OpenAI, due to insane prices and demonic policies). Griffin is already made on EleutherAI, and Dragon is going to have a similar treatment.
On the side note, about 3 months ago, AI's output became much better then it was before, and in a recent days, it is getting even better. My stories contain vore pretty much all the time (mostly safe), and if you can describe it at least somewhat, the AI will roll with it quite nicely. You should give it a try, again.

[AlluringPredation]

What LostCause says above I agree with completely, their understanding of the situation accurate.

I'm going to say there is no redemption. While it's possible that AID's new provider could be used for vore in entirely sfw consensual contexts, as soon as a lewd keyword is struck or non-consensual words are detected the user will be kicked to GPT-J, the same AI model NAI and Holo use. You're still choosing between three groups using the same AI model, two of which have specialized in making the most out of the 6Billion parameter limitation, and Latitude using it as a backup. Considering that NAI's finetune was last stated at being 5.6GB while Latitude's last known finetune size was 30 MB shows Latitude relies completely on the strength on provider's AI.

It's also important to keep in mind that Latitiude, OpenAI, and their new provider, AI21, ALL have major privacy concerns. In that NOTHING you do is private, your EVERY action is stored on each of their servers and is accessible even after deletion requests. While NAI and Holo's privacy is the best you could ever achieve, aside from phrase biasing, nothing is stored server-side unless enable and will be permanently deleted at the click of a button. Including Phrase biasing.

Ecological footprint is an important factor, the processing power required to use AI is exponential in relation to size. 6B-Jax CAN be run on a beefy consumer graphics card and incredibly comfortably on a single A100 (with GPT-Thicc20B possibly as well), while the 175+B classes of AI require entire server entire rack(s) of the same. It's like the difference between taking the bus to work and driving a semi.

As it currently stands I stopped using Latitude's dragon when it's comprehension and output abilities dropped below NAI-Sigurdv3 (currently at v4). Maybe it's better now, but that's in spite of Latitude, not because of them. It should never have dipped to that point, never should a 175B AI model EVER be outperformed by a 6B. It's insanity. It's David vs. Goliath.

[AlluringPredation]

Fun times during testing. Sometimes nonsensical but a good amount of useable material. Plenty of retries here as was testing new LB functions combined with regex, and did hidden subtle guiding with sentence snippets thrown into phrase biasing with completion enabled. Can see my banned tokens/badwords, that will completely derail a vore scene. Some common words/tokens debiased that aren't seen here. Like blood/bloody, as they do have common uses I'd rather not eliminate completely.

[SouthArrival]

Edit: wrong thread, sorry.