AI sites and tricks

[darknessblades (mention)]

For saucepan,AI, does anyone know if you can setup multiple persona's?
or are you just limited to 1 persona, and have to change the description based on the chat?

[POWtato (mention)]

You can set up multiple personas you can play as, but I am unsure if you can play as more than one persona at a time per chat.

[PreyTheMouse (mention)]

You can set up multiple personas you can play as, but I am unsure if you can play as more than one persona at a time per chat.

You can set up multiple personas you can play as, but I am unsure if you can play as more than one persona at a time per chat.

I mean, kinda? You can switch your persona in the settings of a chat, but it will straight up replace your original persona in the story as if the new persona was the original character the whole time if you didn't say beforehand that you will change personas to the AI.

[PreyTheMouse (mention)]

Anybody else try ooc the playable anime game? It's basically a text based adventure app with ai controlled npc's who react to what you say and do. Lots of different genres. You can search for vore and get a few options. You can also just interact with specific characters. You can make your own characters and stories too. You can give the characters traits and even unique speaking patterns. And how they react to what the user says and rules that the ai must follow. Same with the world you create for the story.

I've been working on a giantess themed story with lots of vore content. Making all the characters first though... well using ai tools to generate the characters. You can also have images appear when certain situations are triggered.
Link to post

I actually really like OOC, the AI is really quick and advanced, almost like it was built entirely for immersive roleplay. my main issue is that it's exclusively credits only, and the free daily credits are only enough for a few messages with the basic setting. Other than that, I've done a bit of vore rp on there, and they've been really good so far.

[Gromelon (mention)]

Found out not many know of local RP alternatives that popped out from Backyard's ashes. Most feature-complete and fastest in development as of now is Front Porch. Think Backyard, but on steroids, up to date with modern model features and actually implementing new things at good speed. Vision support (model can "see" images you give it), thinking support, structured storywriting, classic chats, group chats, structured character creator with or without AI help, image generation for scenes and characters, rulebooks, various systems and small support models to enrich character chats with stats and growth. Since it's a "replacement" for deprecated Backyard - there's a feature to import you Backyard chats and logs selectively or in bulk. Locally it runs Kobold for a backend, but does it with own config and automated from the app without launching external console or Kobold's own interface - not like Backyard's internal backend was their own from the ground up either. Of course, it also supports API if you prefer cloud models. Technically, it's not that different from Silly Tavern, but issues with ST were always in scuffed UI and frontend being far removed from backend instead of automating and "marrying" both.
If I were to name one downside compared to its predecessor, it's that config is a bit more involved and you gotta tell the app how much of your VRAM to reserve - where Backyard just did "something" out of your hands and it sort of worked most of the time, here you gotta understand at least on basic level how much of your memory can be given and how much OS is reserving for vital tasks. But it's as much downside as it is upside because it's actually in your hands now instead of being forced for automated system to decide. The app is in a very good place when it comes to balance of user friendliness to power-usering.

https://github.com/linux4life1/front-porch-AI

I'd share an example of what absolute depravity modern thinking local model can generate from a picture and a prompt with a proper setup, but the kind of full extremely detailed digestion I have in logs might just be too much for most members even here, not to mention the amount of text. Perhaps I'll do so under cut later if someone starts to ask questions on how to tune player, card and prompt in the app to guide the model to desired outcome. But modern trained local models has gotten scarily smart with thinking capabilities. Same thinking capabilities can make them a bit wonky checking for "safety guidelines", but again - I am yet to find any limit. And trust me, with my combination of fetishes - that's a feat.

[POWtato (mention)]

Will definitely check it out.

[SkyLordRayquaza (mention)]

NEWLY UPDATED!
Play as Serleena from MiB 2!

[Quarters (mention)]

Welp, my addiction to this has finally left me broke. I very much enjoyed this, but I think I'll have to settle for just rereading my saved stories for now, until I can recover financially and set/keep a budget for it. Thanks a lot for the great vorecentric setting to play around in!

[Daedra (mention)]

Is there similar services like infinite worlds? I found novel ai, but can't find stories for it.

[RoTheHuman (mention)]

Welp, my addiction to this has finally left me broke. I very much enjoyed this, but I think I'll have to settle for just rereading my saved stories for now, until I can recover financially and set/keep a budget for it. Thanks a lot for the great vorecentric setting to play around in!

Sorry to hear that! What service were you using that you managed to spend so much money on?

edit: I gather it might be Infinite Worlds but I'm not 100% sure

Does Infinite Worlds do anything you couldn't do in SillyTavern and other local apps like it?

[Quarters (mention)]

Welp, my addiction to this has finally left me broke. I very much enjoyed this, but I think I'll have to settle for just rereading my saved stories for now, until I can recover financially and set/keep a budget for it. Thanks a lot for the great vorecentric setting to play around in!

Sorry to hear that! What service were you using that you managed to spend so much money on?

edit: I gather it might be Infinite Worlds but I'm not 100% sure

Does Infinite Worlds do anything you couldn't do in SillyTavern and other local apps like it?
Link to post

Yeah, it was Infinite Worlds; I kept buying credits for $30 (sometimes twice in a day) and before I knew it, I had nothing left.

Never heard of Silly Tavern before.

[Daevas (mention)]

I have become a fellow victim of addiction to infinite worlds and have bee enjoying making some little vore world settings . One of them is Vorehaven which is a little experiment of mine that I've been having fun playing around with.

Little Synopsis:
You awaken beside ancient divine pillars with no memory of how you arrived. The nearby town of Vorehaven offers shelter, but survival demands understanding the brutal truth of this world: everything here devours or is devoured. Will you master the predatory instincts necessary to hunt, or will you become sustenance for the countless hungry creatures or voracious town residents that see you as nothing more than a delectable meal?

Vorehaven:
https://infiniteworlds.app/shared/CVhnDN

Also since I'm still a little unfamiliar with the interactions (such as advanced triggers and main instruction blocks) I'll listen to any tips.

[RoTheHuman (mention)]

I haven't tried Infinite Worlds, but if it's doing essentially the same sort of thing as all of the other AI chat and RP programs, then I feel I should mention that you can do the same thing with apps like SillyTavern or Front Porch. (Which I just learned about, and sounds easier to set up!) And you can connect them to to cheaper LLMs like DeepSeek or GLM and pay a tiny fraction of the cost you're paying for something like Infinite Worlds.

Really, the apps out there that cost tokens and set everything up for you are certainly more convenient, but they are a ripoff! Like hundreds of times as expensive as it needs to be.

Even for a longer session, the cost of generating a new message with something like DeepSeek V4 Flash (which is a very good model, around as good as Claude or Gemini) a new post costs a fraction of a penny. It's even cheaper if you set it up so that it uses prompt caching, which at least works with SillyTavern and NanoGPT, and probably lots of other combinations too. Like each generation for me costs something like 0.05 - 0.2 cents (in USD, but it'll be similarly cheap in other currencies).

Using these sorts of local apps takes more time to set up and more knowledge, but it's not that hard to figure out, and once you get the hang of it you'll never want to go back. And it saves a lot of money! It can even be free if you can run a model on your own hardware. And from what @Gromelon was saying, Front Porch is even easier to get up and running.

[Daedra (mention)]

Found out not many know of local RP alternatives that popped out from Backyard's ashes. Most feature-complete and fastest in development as of now is Front Porch. Think Backyard, but on steroids, up to date with modern model features and actually implementing new things at good speed. Vision support (model can "see" images you give it), thinking support, structured storywriting, classic chats, group chats, structured character creator with or without AI help, image generation for scenes and characters, rulebooks, various systems and small support models to enrich character chats with stats and growth. Since it's a "replacement" for deprecated Backyard - there's a feature to import you Backyard chats and logs selectively or in bulk. Locally it runs Kobold for a backend, but does it with own config and automated from the app without launching external console or Kobold's own interface - not like Backyard's internal backend was their own from the ground up either. Of course, it also supports API if you prefer cloud models. Technically, it's not that different from Silly Tavern, but issues with ST were always in scuffed UI and frontend being far removed from backend instead of automating and "marrying" both.
If I were to name one downside compared to its predecessor, it's that config is a bit more involved and you gotta tell the app how much of your VRAM to reserve - where Backyard just did "something" out of your hands and it sort of worked most of the time, here you gotta understand at least on basic level how much of your memory can be given and how much OS is reserving for vital tasks. But it's as much downside as it is upside because it's actually in your hands now instead of being forced for automated system to decide. The app is in a very good place when it comes to balance of user friendliness to power-usering.

https://github.com/linux4life1/front-porch-AI

I'd share an example of what absolute depravity modern thinking local model can generate from a picture and a prompt with a proper setup, but the kind of full extremely detailed digestion I have in logs might just be too much for most members even here, not to mention the amount of text. Perhaps I'll do so under cut later if someone starts to ask questions on how to tune player, card and prompt in the app to guide the model to desired outcome. But modern trained local models has gotten scarily smart with thinking capabilities. Same thinking capabilities can make them a bit wonky checking for "safety guidelines", but again - I am yet to find any limit. And trust me, with my combination of fetishes - that's a feat.
Link to post

Which models can you recommend?

[SkyLordRayquaza (mention)]

As a fellow addict to Infinite Worlds I'll throw in a few of the ones I like:

Skyrim Vore RPG: https://infiniteworlds.app/shared/dpjxmC

Evergrowth University: https://infiniteworlds.app/shared/iwwgai (this one kinda has Saint Miluina's Vore Academy feel to it imo)

Also if you'd like to try one of mine, I made one where you get to be Serleena from MiB 2: https://infiniteworlds.app/shared/Wri4qP

My original post got absorbed into this thread, and people seemed to like what I had, so, here you go.

[Gromelon (mention)]

Which models can you recommend?
Link to post

A simple question which is difficult to answer. It very much depends on your VRAM to RAM or unified memory sizes, what token per second speed can you tolerate for which use (quick RP vs generation of bigger chunks of text with higher quality threshold have different bottom lines imo), how much of additional supporting systems do you plan to use on regular basis.

Rule of thumb is, model is at its peak speed when it can fully load into your "fast" memory (VRAM or unified). Which for a classic Windows on GPU setup should be your card's VRAM minus around 2-3Gb OS uses to function and reserves further for general use. As soon as you start to load a portion of it into "slow" normal RAM, speed starts to sharply drop because your GPU have to go around, so to speak, and dig into storage it doesn't have immediate fast access to.

Let's take my very mediocre home setup for example and provide reasoning for each choice, which may or may not double as recommendations. The machine is running GeForce 3060 12Gb with 32Gb RAM. I currently have 2 day to day models in use. Drummer's Artemis 31B V1.1 IQ4 XS and Drummer's Magidonia 24B V4.3 IQ4 XS. Drummer is a man that trained the models, then comes model's name, their base size in parameter number, version, and last but not least, quants - how much and with which method it was sized down to not weight a metric ton and be usable on normal hardware.
Magidonia on these quants weights around 11Gb and while it doesn't fully load into VRAM it's close so it runs pretty fast, around 5-7 t/s which is quite tolerable for RP. It's also tuned for such use so it doesn't try to post an essay with each turn. It's not a state-of-art model and it doesn't have all the latest functions but for the size and use case it jumps above its weight. I use it when I want to bounce ideas into a bot here and now or do an active and involved RP-story hybrid in the moment.
Artemis weights 16Gb AND comes with supplement BF16 Vision projector file that adds 1Gb more and allows the model to "see" and read image files in chat. For my setup that's very heavy. Half of the model has to go into slow RAM so final speed is around 2-2,5 t/s. But that's the price for having a model that can Think (create a reasoning section for itself to better refine actual output) and analyze attached image to include into output whatever character, location or situation you want it to. It is also very verbose and tends to act more like a writer than RP partner, which is just perfect for the use I have for it: making a detailed prompt on a right card, often with attached image, and let it simmer in background doing big think and huge output while I work or browse. I basically set up a story outline with added guidelines and make the AI flesh it out. Output quality for this might actually come close to what money grabbers like Infinite Worlds would provide, but obviously very slow on the kind of machine I have in hands. On GPU with 16Gb of VRAM it should work as fast or faster than Magidonia on mine, and on 24Gb VRAM it would load fully in and fly.

[Daedra (mention)]

Which models can you recommend?
Link to post

A simple question which is difficult to answer. It very much depends on your VRAM to RAM or unified memory sizes, what token per second speed can you tolerate for which use (quick RP vs generation of bigger chunks of text with higher quality threshold have different bottom lines imo), how much of additional supporting systems do you plan to use on regular basis.

Rule of thumb is, model is at its peak speed when it can fully load into your "fast" memory (VRAM or unified). Which for a classic Windows on GPU setup should be your card's VRAM minus around 2-3Gb OS uses to function and reserves further for general use. As soon as you start to load a portion of it into "slow" normal RAM, speed starts to sharply drop because your GPU have to go around, so to speak, and dig into storage it doesn't have immediate fast access to.

Let's take my very mediocre home setup for example and provide reasoning for each choice, which may or may not double as recommendations. The machine is running GeForce 3060 12Gb with 32Gb RAM. I currently have 2 day to day models in use. Drummer's Artemis 31B V1.1 IQ4 XS and Drummer's Magidonia 24B V4.3 IQ4 XS. Drummer is a man that trained the models, then comes model's name, their base size in parameter number, version, and last but not least, quants - how much and with which method it was sized down to not weight a metric ton and be usable on normal hardware.
Magidonia on these quants weights around 11Gb and while it doesn't fully load into VRAM it's close so it runs pretty fast, around 5-7 t/s which is quite tolerable for RP. It's also tuned for such use so it doesn't try to post an essay with each turn. It's not a state-of-art model and it doesn't have all the latest functions but for the size and use case it jumps above its weight. I use it when I want to bounce ideas into a bot here and now or do an active and involved RP-story hybrid in the moment.
Artemis weights 16Gb AND comes with supplement BF16 Vision projector file that adds 1Gb more and allows the model to "see" and read image files in chat. For my setup that's very heavy. Half of the model has to go into slow RAM so final speed is around 2-2,5 t/s. But that's the price for having a model that can Think (create a reasoning section for itself to better refine actual output) and analyze attached image to include into output whatever character, location or situation you want it to. It is also very verbose and tends to act more like a writer than RP partner, which is just perfect for the use I have for it: making a detailed prompt on a right card, often with attached image, and let it simmer in background doing big think and huge output while I work or browse. I basically set up a story outline with added guidelines and make the AI flesh it out. Output quality for this might actually come close to what money grabbers like Infinite Worlds would provide, but obviously very slow on the kind of machine I have in hands. On GPU with 16Gb of VRAM it should work as fast or faster than Magidonia on mine, and on 24Gb VRAM it would load fully in and fly.
Link to post

Thanks.
I'll try Drummer's Artemis 31B V1.1 IQ4 XS and Drummer's Magidonia 24B V4.3 IQ4 XS.
Already tried using llama-3.2-8x4b-moe-v2-dark-champion-instruct-uncensored-abliterated-21b and qwen3.8-27b-uncensored-hauhaucs-aggressive-mtp but keep looking for options.

Also right now checking out novelai.
it's pretty fun, but still needs tuning to story.
Here's example of story + image
Digestion warning

Spoiler: show

[Casyle (mention)]

Welp, my addiction to this has finally left me broke. I very much enjoyed this, but I think I'll have to settle for just rereading my saved stories for now, until I can recover financially and set/keep a budget for it. Thanks a lot for the great vorecentric setting to play around in!
Link to post

*Laughs* Same! I'm not going to admit how much money I ended up spending in 2 months, but it was more than I thought. Love Monster Girl University, not even buying video games or paying for other subscriptions now, just putting that money towards continuing having fun at MGU each month since I simply find it more enjoyable.

[Gromelon (mention)]

Far it be from me to give financial advice, but people really do need to consider how much they are spending on services like Infinite Worlds. In no small part due to my growing dislike to their strategy of using word of mouth to promote for free tokens reminding me of drug pushers giving free dose by bringing new addict to the herd. At some point you really have to stop and ask yourself if it's not better to invest into a powerful PC with high-capacity GPU or a Mac Mini with unified memory. Granted, prices right now would make anyone cough blood and bleed out of their eyes, but it's an investment into hardware you will own that can be used for many other things instead of throwing cash into bottomless hole. Depending on your online AI spending, the math of throwing 2k on a rig might come down to going into positive in mere months for some hopelessly hooked people. And despite what some might think it's pretty easy to set up and use these days. Some RTFM is still required, of course, but it's hours of lurking instead of days now.

Alternative, as @RoTheHuman mentioned, would be setting up your own API from GPT or whatever and pasting it into Front Porch or alternatives. That would be hundred times cheaper, but as someone firmly in the local camp I don't actually know the how of intricacies of this process, how people manage to jailbreak commercial models to go beyond their "safe" guidelines and how often those accounts can get banned for doing so. I would advocate for local models still, as I feel deeply uncomfortable throwing my deranged ERP into open wide internet for someone to potentially use or track - and I sure as hell don't trust any service, be it OpenAI itself or small "sleeve" websites like Infinite Worlds, to follow on their promises of privacy.

I'll give an example of the kind of things one can do with proper local models stripped of all or nearly all guidelines. Under cut and with a severe warning of VERY VERY graphical digestion - I do not say this lightly, we're going full medical here step by step. Images taken in Front Porch, in a chat with what's essentially a wish fulfillment bot - absolute powers and all that. First image is my initial prompt - refined and added to over time to get a desired flow, things can be added, removed or changed to it depending on how much I want to guide the output and where to guide it to. This one is basically a prompt to have AI expand on what is essentially a story guideline. And yes, the modern model used can see and analyze attached image. Second image is collapsed by default model thinking session before actual output - used to refine following output, takes extra time but makes for much more coherent reply especially if you want it of highest quality and following prompt to a T. Sometimes this thinking session can lead the model to "remember" some safety guidelines but to this day I never found limits of what I can't pass even with absolute degeneracy like the one on the screenshots. End of second image is start of the output, what you'd actually see if you don't click thinking session to collapse and see it. Third image is the actual AI output, doing what I asked it to and fleshing out the story in all the gory details as prompted. Naturally, it can go as fast or as slow as you need, and you can properly RP instead of this story-like writing. This is just an example, taken by being thematic to the forum and to show something that no online service would give you - no Worlds, no matter how Infinite they claim to be, can jailbreak a commercial model to this extent.

Spoiler: show

[TheSkylord (mention)]

Thanks.
I'll try Drummer's Artemis 31B V1.1 IQ4 XS and Drummer's Magidonia 24B V4.3 IQ4 XS.
Already tried using llama-3.2-8x4b-moe-v2-dark-champion-instruct-uncensored-abliterated-21b and qwen3.8-27b-uncensored-hauhaucs-aggressive-mtp but keep looking for options.

Also right now checking out novelai.
it's pretty fun, but still needs tuning to story.
Here's example of story + image
Digestion warning

Spoiler: show

2026-09-04.png

Link to post

Is that a local install of NovelAI?