Unofficial NovelAI General

Keep our community informed! This forum is for discussing and sharing vore-related information. Post any relevant material and/or links here, and engage in conversations!
Forum rules
This is for general discussion, if you found something you want to post, please use one of the upload forum, if you made something and want to share them, please use the work to be shared forum!

Re: Unofficial NovelAI General

Postby LostCause » Wed Nov 24, 2021 7:41 am

Interesting. Thank you for the responses!
User avatar
LostCause
Participator
 
Posts: 274
Joined: Sat Feb 20, 2021 8:55 am

Re: Unofficial NovelAI General

Postby AlluringPredation » Wed Nov 24, 2021 3:49 pm

I've compiled some regex for LoreBooks that might come in handy. And because I want them somewhere for own personal reference. Of course you CAN use the basic feature for keyword entry, but that's not fun.

/keywords?/i - Search for a keyword where a plural is optional; keyword and keywords. / and / are required to state that this is regex, the i following the / states that you want the keywords to be case insensitive; Keyword, keywords, or even KEyWoRDs. The ? denotes the character prior is optional, so accounts for plural forms.

/(keyword|keywords?)\b/i - By putting your words in parenthesis and separating them with a bar | you can say either this keyword or that. Generally if you keyword: breast WILL match on (breast)s but by placing a word bounday with `\b` you keep it from triggering on breastplate and still account for pluralization.

/fox-?\s?(girl|woman)/i - This is for monstergirl matching stuff. Fox's specifically obviously, the ? makes the previous character optional (dash and (s)pace) enabling it to match a variety; Foxgirl, Fox-girl, Fox girl; and woman in the same ways. Fox easily changed to anything, and making it case (i)nsensitive, make capitalization not matter. Recommend either using a - or space between. The way words are broken up can lead to issues, like dragongirl is ready by the ai as `drag|on|girl` and fixates on `drag` and `girl` becoming a story about crossdressing.

/^((?!Keyword).)*$/i or /^((?!Keyword|Keywords).)*$/i - This searches context for LACK of associated keyword, regex is limited and doesn't work the best for this; Will be a lot more powerful if category keywords are created, or through scripting. Still you can search for lack of dialogue indicators and pronouns to trigger random encounters with Lorebook/PhraseBias.

/(?<=Tags:).*keyword/i - Using this one along with Cascading activation to search the Author's note for specified keywords, combined with LB phrase biasing to boost/negate certain words to push the narrative in the desire direction. Currently have vore/ingest/swallow/digest/absorb for phrase biasing, but they do little as Author's note tags yet, will be trying to tag certain events in my module data to create an effect like I have with ' vore'

/(.*?\b(Keyword1?|Keywords?)\b){2,}/is - This looks for keywords but also requires they appear x amount of times currently set at {2}. This can make a Lorebook a little less sensitive by making it not appear at a single mention. OR, if it's mentioned more often you can provide an more context through additional lorebooks. [ NAI: You sure are mentioning digestion a lot, let's go about ending you and how your preds feels afterward.]

/(Tags:).{0,40}(TagKeyword)(.*?\b(Keyword1?|Keywords?)\b){1,}/is - This combines two above, checks Author's Note for certain topics AND checks to see if these keywords appear later in the context, also able to set a requirement for how many times they must appear (set at 1). Could very well enable making sure the right LoreBooks are applied at the right time. This can get resource heavy, and I'm staying away from having it look before `Tags:` as that might create some heavy compute issues. If we get category searching this would be very powerful. (Note the {0,40} exists so NAI can search up a certain amount without taking too much context into keyword searching, might improve this one later.)

/keyword.*((?=[a-n]).{10})$/si + /keyword.*((?=[^a-n]).{10})$/si - This is a fun one for shuffling multiple lorebooks with the same keyword and randomly using either. It counts back from the end of context, currently {10} (can adjust), spaces to see what letter appears. The first one {a-n} will match on letters a-n, the second will match anything NOT (denoted by ^) a-n (including spaces/punctuation/whatever.)

/keyword.*((?=[a-i]).{10})$/si + /keyword.*((?=[j-p]).{10})$/si + /keyword.*((?=[^a-p]).{10})$/si - Same as above but 3 possibilities, you can always do more. Most importantly the last one with the ^NOT must always be all the letters you are searching for to catch ALL OTHER POSSIBILITIES. So you can adjust the letters as you wish to change probability distribution. I'll be using this to shuffle lorebooks around to keep descriptions from becoming stale. Just make sure to use other numbers besides just {10} for maximum randomness. It's not TRUE randomness but does take advantage of the ever-changing story context to keep it fresh.
Discord for artists and those who commission them! https://discord.com/invite/F3UDqqAx53
User avatar
AlluringPredation
Participator
 
Posts: 293
Joined: Fri Mar 27, 2015 11:10 am

Re: Unofficial NovelAI General

Postby Redley » Sun Jan 09, 2022 10:01 pm

Just tested NAI's new model, Euterpe, and wow! It seems to be knowing how vore works without additional modules and even with empty lore. My favorite way to test AI's knowledge about something is to make it write "10 facts about X". And this time AI explained what is vore, how it works and who performs it, without random made-up stuff.
Also looks like Euterpe can understand the sequence of vore scene. Teasing, then consumption, then digestion. Even pre-lobotomy Dragon was not capable of it.
Generally the AI feels a bit tongue-tied for now, just like Sigurd on release, but the logical part is looking so good I can barely believe it.
User avatar
Redley
Been posting for a bit
 
Posts: 56
Joined: Mon Jun 08, 2009 9:09 am

Re: Unofficial NovelAI General

Postby AlluringPredation » Mon Jan 10, 2022 6:49 pm

Yeah, it's hellaswag (reading comprehension) test scored really well for its size, and it shows by following action pretty well. Did some tests myself and it displays an understanding of many more concepts than Sigurd does.

Still early on at the first fine-tune only at 65%. I've noticed a habit of occasional repetition so been making sure to active sloping repetition penalty.
Discord for artists and those who commission them! https://discord.com/invite/F3UDqqAx53
User avatar
AlluringPredation
Participator
 
Posts: 293
Joined: Fri Mar 27, 2015 11:10 am

Re: Unofficial NovelAI General

Postby LostCause » Tue Jan 11, 2022 11:22 am

I can't try the model myself at the moment so this is great to hear. Fantastic to hear that it follows action well. Does it keep up with the current situation much better, particularly in vore? I'm also curious whether the new model seems to have better zero-shot/few-shot capability, or whether it feels more like the AI 'understands' more concepts to begin with. What do people that have tried it think? I feel that this will be the most important advancement whenever it comes to vore, since there will never be many examples for it to draw from in the training data and the context is only so big making it difficult to provide it with examples.

Anyone think it's worth using over Sigurd already for vore? Or would it be better to hold off and wait till it's more complete/comes to the cheaper subscription tier? Personally I'm not able to afford a subscription at the moment and I'm not sure whether to go for Opus when I can or opt for the cheaper fee.

I had a bit of a funny thought. I wonder if a model's efficacy for vore could actually be *reduced* by the finetune in a way. Vore and sex are often mixed up so finetuning on coom and romance could raise the prevalence of these mistakes haha.
User avatar
LostCause
Participator
 
Posts: 274
Joined: Sat Feb 20, 2021 8:55 am

Re: Unofficial NovelAI General

Postby AlluringPredation » Tue Jan 11, 2022 1:06 pm

I always felt that Siggy was a few shot model needing modules to make one shots even possible for our weird interests. Euterpe has far better one-zero shot capability, some solid lorebook entries to prime should be enough to get most things to work.

Attempted a mix two uncommon concepts in this example; genderbending and magical girl transformation, and for zero-shot I think Euterpe performed respectably:
[input] He began to undergo a transformation, genderbending form his male form into a magical girl. Sparkles
[output] of magic radiated out from the area where her breasts were supposed to be as he became more and more feminine. Her hair started growing long and curly. A large pair of pink glasses was placed on top of her head and she began to wear an over-sized sailor uniform with frills in front that made it look like a dress. She looked at herself in a full length mirror. "Awww man... I'm too pretty for my own good..."


Euterpe does keep up with the situation better, and can follow the entire active context to a fault; looping back to repeat early story events or latching onto grammatical/punctuation errors. Unless you got money to burn better to hold off for now on getting Euterpe for now, current state feels like it traded some descriptive ability for much better dialogue, I want it all! Euterpe does know OF vore, it can dialogue FOR vore, but DESCRIBING vore is a greater challenge. The Vore/Romance/Sex confusion will likely be helped with finetuning to focus lewd where appropriate, but the overlap needs accounted for with some well-placed lorebooks.

If Euterpe gets into a good state for vore I've considered hosting a stream where people can witness it in action, but that might be some months away.
Discord for artists and those who commission them! https://discord.com/invite/F3UDqqAx53
User avatar
AlluringPredation
Participator
 
Posts: 293
Joined: Fri Mar 27, 2015 11:10 am

Re: Unofficial NovelAI General

Postby LostCause » Tue Jan 11, 2022 3:02 pm

Interesting. It's unfortunate that following the context better causes looping and errors to be more prevalent. Fingers crossed they can curtail that without requiring a hefty repetition penalty.

Shame it isn't great about description (yet?), definitely going to hold off for now then. Hopefully that will change when the finetune is complete because that's the more important part for me. It's great to hear that dialogue is better though, as even after v4 it was mostly nonsense for me if I didn't keep it on a very short leash.

Enjoy tinkering around. Fingers crossed a lot of your work with Sigurd transfers over well enough.
User avatar
LostCause
Participator
 
Posts: 274
Joined: Sat Feb 20, 2021 8:55 am

Re: Unofficial NovelAI General

Postby AlluringPredation » Tue Jan 11, 2022 4:40 pm

They took my suggestion of enabling specific symbols on an empty line to prevent the AI from looking back any further into older context, and that led into discussion of creating a slider for amount of active story context. AID's context was only 600-700 tokens when it was good, and while Sigurd had a 2k context I only felt it was effective at using half. So Euterpe's ability to fully utilize the entire context is an entirely new experience. Might help too with future finetunes giving the AI options in how to continue a story.

It's fun testing things out, each model has some quirks and power words to utilize, but can't know exactly how it will react until we get a more finalized finetune. What works today may not be effective tomorrow in terms of language.

My latest tinkering is using key-relative to insert adjectives to a specific Noun. Basically attaching personality and traits to a character or species;
-'The dragon woman' is adjusted into context as 'The [ towering, crimson-scaled ] dragon girl'

For specific characters I'm trying some regex that looks for newline or end of sentence beginning with a name;
- 'Erica' could become '[ The plump, haughty dragon girl ] Erica'

With the old dragon this was an extremely effective method to reinforce characters and tie them in better with their summaries, hoping to mimic this effect with Euterpe, didn't work so great with Sigurd.

Also Eleuther's 20b is getting close to ending it's training runs.
Discord for artists and those who commission them! https://discord.com/invite/F3UDqqAx53
User avatar
AlluringPredation
Participator
 
Posts: 293
Joined: Fri Mar 27, 2015 11:10 am

Re: Unofficial NovelAI General

Postby LostCause » Tue Jan 11, 2022 5:23 pm

You didn't have to wait long because they already finished the first fully trained version lol.

Throwing out a random idea that may or may not be useful: Perhaps that bracket method won't work so well with a larger context size? It might have worked better with Dragon because it wouldn't see so many instances of it in the context? Though I suppose in reality it'd more likely be because Dragon could better keep up with characters without near-constant name drops tied to their words and actions.
User avatar
LostCause
Participator
 
Posts: 274
Joined: Sat Feb 20, 2021 8:55 am

Re: Unofficial NovelAI General

Postby AlluringPredation » Tue Jan 11, 2022 7:37 pm

It works reasonably well, in old AID it was too powerful and had to adjust stuff with the borrowed script to reduce it's effects.

I'm going to be in the minority and say I'm disappointed in Euterpe. I feel like vivid descriptions were traded for dialogue and we end up with too much 'tell' and now enough 'show' in the output even when filling the context with a decently written example to springboard from. This may satisfy many, not the storytelling style that satisfies me. This may change in future finetunes, I remember that it took until Sigurd v3 until I felt it was minimally useable and v4 can provide some nice adjective use that helps with my weak areas.

I'll probably return to waiting for 20b and hoping it Eleuther can manage to pull out a win with it's model. I feel like the addition of clinical knowledge helps for the consumption aspects of vore that 13b lacks.
Discord for artists and those who commission them! https://discord.com/invite/F3UDqqAx53
User avatar
AlluringPredation
Participator
 
Posts: 293
Joined: Fri Mar 27, 2015 11:10 am

Re: Unofficial NovelAI General

Postby LostCause » Wed Jan 12, 2022 6:11 am

That's a shame. Perhaps when modules are more developed or a further finetune is complete it'd reach the level you wish for. Sigurd has never really provided vivid descriptions (for vore) in my experience, except on rare occasions out of the blue, so at least in that case it wouldn't make so much difference to me. Any attempt to get more fancy or descriptive only seems to 'confuse' the AI, most of the time I have to keep it more simple and direct. Could be a result of my lacking writing skills and sentence structure though.
User avatar
LostCause
Participator
 
Posts: 274
Joined: Sat Feb 20, 2021 8:55 am

Re: Unofficial NovelAI General

Postby LostCause » Wed Feb 09, 2022 1:54 pm

In all my usage of Euterpe v1 and v2 so far I've noticed that I cannot for the life of me get head-first (same size) swallowing scenes to work well. Feet-first on the other hand seems to be pretty consistent once past the initial foot fetishism barrier. Has anyone else noticed this or is it just me? What I find most odd is that the main problem isn't kissing or sex related outputs as was often the case previously. It simply appears to devolve into nonsense most of the time. Even when I write most of the scene myself or make tons of edits, maintaining coherence for the rest of the story afterwards is a real challenge, it appears to really 'throw off' the AI. Also lots of outputs about swallowing ONLY the head and ignoring everything else—as if the training data has tons of material about biting heads off then swallowing them whole or something odd like that???

Anyone else observed something similar to this?

Apart from that I've mostly been trying to find the right preset(s) for me. What presets or sampling settings have you guys felt worked best with vore so far? Before Euterpe v2 came around I personally had settled on using Mothra v3 mostly, not sure yet when it comes to v2.
User avatar
LostCause
Participator
 
Posts: 274
Joined: Sat Feb 20, 2021 8:55 am

Re: Unofficial NovelAI General

Postby AlluringPredation » Fri Feb 18, 2022 8:20 pm

Euterpe is terrible for vore, my experiences mirror yours perfectly. Even with plentiful examples in LoreBook, my module full of personal story examples, and boatload of bias settings...At best it just; I'm gonna eat you, I ate you, pat belly. Meanwhile the same settings in Sigurd have better results. I hit cancel on my subscription until 20B hits.

Fairseq 13B, Euterpe's base funded by Facebook/Meta, uses a much smaller dataset in its training, and reviewed it 3 times over with a heavy focus on chatting. Vore not included. I really honestly DO NOT like it. Even when it works I really do not care for the dialogue, often stuff you've heard 100 times before online and doesn't feel tailored to the situation, and I hate the writing style it tends towards. This happened, that happened; Rather than vivid descriptions.

20B will be using the same dataset as Sigurd, with roughly the same level of comprehension according to tests. It does have a far greater score for medical knowledge, which I believe will be a great boon to us in having the AI understand the inner workings of the body far better. Even testing the raw/unfinetuned version I can see hints of good results. You can try it out @ https://goose.ai/playground, just don't expect coherence. You get $10 credit during a trial, which lasts quite a while.

This is from unfinetuned 20B;
Input:
This is a vore story, where people are swallowed alive by alluring monstergirls.
Detailed description of a man being swallowed alive by a lamia: His wriggling body passing through her lips one inch at a time beginning with the man's feet being introduced into the drooling, dark maw of the snake woman. He

Output:
could feel her mouth opening more and he was aware that her eyes were focused on him. His vision blurred as the snake girl's saliva smeared around his face. His world fading as his body was drawn into her yawning gullet.
The man's head, torso and legs disappeared inside her cavernous body, and he could not stop himself from shouting in protest.


While the results are not fantastic, nor should the expect to be with a raw AI model (took a couple retries to get this result), it does show hints of vorey goodness. The concepts are there, just needs a proper education through finetuning on how stories progress.
Discord for artists and those who commission them! https://discord.com/invite/F3UDqqAx53
User avatar
AlluringPredation
Participator
 
Posts: 293
Joined: Fri Mar 27, 2015 11:10 am

Re: Unofficial NovelAI General

Postby LostCause » Sun Feb 20, 2022 5:00 pm

After more time spent with Eute I have to say I agree. Even switching to Sigurd mid-story often produced better results than Euterpe would on constant retries. I'm in the same boat of waiting for 20B now, I'm looking forward to what will hopefully be a more intelligent and coherent Sigurd. I'm guessing the NAI finetune dataset has grown and changed significantly since Sigurd v4 was trained, but I'm sure that will not be a detriment to our usage.

Fingers crossed for a surprise 20B v1 release soon haha.
User avatar
LostCause
Participator
 
Posts: 274
Joined: Sat Feb 20, 2021 8:55 am

Re: Unofficial NovelAI General

Postby AlluringPredation » Sun Feb 20, 2022 6:17 pm

*Soon, but might be a bit for 20B. Discussion appears they may be trying to find optimizations enough to get it utilize full 2048 token context with all the trimmings on a single 48GB GPU rather than using Dual, which is a big cost savings and speed boost. It's also a massive undertaking, as it's said they're the only group able to get 20B up and running on a 48GB GPU. The continued sky-high cost of computer chips and GPUs in turn is the largest obstacle for this technology. Their fine-tune also sits about 5GB, which adds quality but RAM overhead as well.

Improved finetune shouldn't present an issue, that little bit of output I gave shows that 20B AI has the info somewhere, even if not specifically trained. Also there may be some vore in the finetune, even if just a fraction as Sigurd v4 had better understanding than Sigurd v3. Or someone could write the most amazing vore story ever and submit it to NAI's finetune team!

I'm letting my NAI subscription lapse, partially because I need to study art. Partially because want to play around rewriting some stories I have laying around. To share, and to insert into a custom NAI module.
Discord for artists and those who commission them! https://discord.com/invite/F3UDqqAx53
User avatar
AlluringPredation
Participator
 
Posts: 293
Joined: Fri Mar 27, 2015 11:10 am

Re: Unofficial NovelAI General

Postby LostCause » Tue Feb 22, 2022 6:02 pm

Interesting stuff. I haven't been keeping up with discussion for a while now so I had no idea about any of that. I wonder if that means they want to try to allow access to all subscription tiers too instead of only Opus. I was surprised when they did that for Euterpe, I fully expected it to stay Opus-exclusive at the very least until 20B came.

Good luck with your projects and thanks for sharing what you heard.
User avatar
LostCause
Participator
 
Posts: 274
Joined: Sat Feb 20, 2021 8:55 am

Re: Unofficial NovelAI General

Postby shaw843 » Sat Mar 05, 2022 6:38 am

AlluringPredation wrote:

=========================================================
[u][b][size=150]Tools of NovelAI

Vore is woefully underrepresented in the training data. We're all aware vore itself goes against the natural process of consumption and breaking many rules of reality. Prepping NovelAI for vore thus presents some challenges. Yet the developers have equipped users tools for the task. Namely; Lorebooks, Phrase Biasing, and Modules. Each requiring an escalating amount of effort to properly utilize

Lorebook
- an absolutely neccessary for vore, and easiest to implement. Lacking proper understanding the AI needs a few examples of vore in Context, giving it instructions in how to proceed.
It is best to playtest repeated examples and keywords combinations that will yield desired results. Feel free to share examples and .Lorebook files to help one another out. But please make sure that the entries contained within a lorebook entry not contain copyrighted snippets of text, as lorebook can sometimes be utilized verbatim and this would constitute plagiarism.

Think you could share those lorebooks?
shaw843
New to the forum
 
Posts: 19
Joined: Fri Oct 23, 2015 11:08 pm

Re: Unofficial NovelAI General

Postby CreativeBunny » Wed Apr 06, 2022 3:37 am

shaw843 wrote:
AlluringPredation wrote:

=========================================================
[u][b][size=150]Tools of NovelAI

Vore is woefully underrepresented in the training data. We're all aware vore itself goes against the natural process of consumption and breaking many rules of reality. Prepping NovelAI for vore thus presents some challenges. Yet the developers have equipped users tools for the task. Namely; Lorebooks, Phrase Biasing, and Modules. Each requiring an escalating amount of effort to properly utilize

Lorebook
- an absolutely neccessary for vore, and easiest to implement. Lacking proper understanding the AI needs a few examples of vore in Context, giving it instructions in how to proceed.
It is best to playtest repeated examples and keywords combinations that will yield desired results. Feel free to share examples and .Lorebook files to help one another out. But please make sure that the entries contained within a lorebook entry not contain copyrighted snippets of text, as lorebook can sometimes be utilized verbatim and this would constitute plagiarism.

Think you could share those lorebooks?


I would be interested in that too. I hear they made lorebooks better too.
User avatar
CreativeBunny
New to the forum
 
Posts: 2
Joined: Wed Apr 06, 2022 3:33 am

Re: Unofficial NovelAI General

Postby WebpagesAreTrees » Tue May 31, 2022 12:41 am

Hello people. I've got a bit of experience with wrangling AI writing assistants for vore purposes. I had probably 100,000 or so actions on AIDugenon's Dragon model before it imploded, and I've got experience with all the NovelAI models. I'm a prolific vore writer, as I've probably written over 500,000+ characters of vore smut just through NovelAI at this point. I've got lots to share in terms of techniques, tips, and resources for vore. I've made several vore models based on other vore writers and on my own, and I've written 40 - 50 vore prompts that are available for public use.

I'm specifically chiming in because I'd like to mention all the wonderful experiences I've had with Krake v2 (NovelAI's latest 20b model) in terms of vore output. For full context, I'm mostly into animal/monster/monster girl predator on human/elf/anthro prey soft vore. I have noticed that the AI struggles with the more esoteric types of vore, like cock/same size/ultra size difference. I have gotten all types of unbirth, and even cock vore to work well, but it can be hit or miss without intensively editing the AI's outputs to keep it on point. With that being said, I'm starting to get incredibly impressed by how coherent the Krake 20b AI model is becoming, as it can maintain coherent utilization of all kinds of variables in vore scenarios. It's revolutionized my experience with vore.

As for advice and strategies, I'm not sure where to begin, as each person's use case is different. AlluringPredation has offered some great advice here that mirrors a lot of my own experiences. Particularly how it is futile to try to teach the AI 'vore' as an abstract concept. What you're really trying to do, is trigger all the information correlations the AI has between predator, prey, and the eating process. The word that universally helps with keeping the AI on track with vore, is 'voracious' and 'predator', describing a character in the memory or a lorebook as a 'voracious predator that enjoys swallowing down their victims whole and alive," is an excellent way to help the AI get the clue. Even describing someone's vagina or cock with predatory/hungry language will help you get good results. 3rd person tense will almost always result in superior results to 1st/2nd person, as there are limited and low quality sources for those types of writings. Another big one, that I may have inured myself to through training, is continuously editing, retrying, and splicing the AI's outputs to keep the story going how you'd like. I feel like I have to put an equal amount of effort into a vore scene as a sex scene now to get good results. If I give the AI too much leash in either scenario it's usually not going to be very interesting or coherent. Once you figure out the specific words that 'summons' the type of AI correlations that lead to vore you enjoy, you can get quite good at autopiloting it, as long as your scenario isn't too abstract. However, the AI can surprise you even with the more abstract stuff if you tune it just right.

I've got lots of screenshot examples of how I 'wrangle' the AI. They are incredibly graphic literary examples, but informative. I can link all the vore prompts I make to give everyone an idea of how I construct a scenario's foundation, or if you just want to jump in without making your own scenario. I can post my modules too, but I don't even use the Euterpe model for vore anymore, as I find Krake (which can't use modules currently) superior. I originally hail from a different NovelAI/AI writing community, so a lot of the stuff I've produced is anchored through them. I'll refrain from oversharing at the moment, as this isn't the forum for posting your work, but I'll be happy to provide any of those things I've mentioned for information purposes. I could also just post a sharing thread in the appropriate sub-forum if that would be the best option.

I'll be checking in if anyone wants to discuss writing with NovelAI or needs advice, happy writing.
User avatar
WebpagesAreTrees
New to the forum
 
Posts: 4
Joined: Wed Jun 08, 2016 11:59 am

Re: Unofficial NovelAI General

Postby noego » Wed Jun 01, 2022 11:24 am

Hmm. I would expect that "Snek" would be better at vore than "Krake." Who knows why! Pondering getting a Novel AI subscription but $25 a month scares me a bit, considering I don't understand this thing a whole lot, but it sure looks alluring as someone that works in programming and would want to write more but is insufficiently creative...
noego
Been posting for a bit
 
Posts: 31
Joined: Tue Dec 27, 2005 12:00 am

PreviousNext

Return to General Vore Discussion

Who is online

Users browsing this forum: No registered users