It REALLY depends on two things :

1- your model of choice.

2- your specifications.

I have 32gb ram (ddr5 @ 6000mhz), rx 6950 xt with 16 gb of vram and an i5-13600k, the model that I got to run comfortably range between 24b and 27b, anything higher than that got my pc screaming for mercy plus slowdowns and incoherent output. With 6 gb of vram, I highly advise you to run Sao10Ks L3 8B Stheno v3.2, either Q3 quants or Q4 quants, but I highly recommend that you download the Q3 quants (for higher token count), assuming that you have a 16gb of ram, you can run between 8k and 12k tokens, which is higher than the token count that the perchance model have (6k tokens), also, use koboldcpp and connect it to sillytavern, it's a bit complicated but once you get it running, it's a very smooth sail.

On a side note, if you could get your hands on a clean card with a lot vram like an rtx 3090 with 24gb of vram on the used market (got one for less than 400 bucks), you can run much, much better models locally, models that would nuke perchance's Ai chat into oblivion with much higher context windows, like Gemma 4 31b, WeirdCompound family, Cydonia v4.1 24b... Etc.

It doesn't matter what you use, OOC is consistent in Ai roleplay, in Ai RPG, Ai chat, sillytavern... Etc and it always worked just fine, but it seems whoever made the story gen had a different idea in how it operates, been fighting the Gen using OOC with 50/50 success rate, it worked and sometimes it doesn't (very hilariously rebellious), I just keep pasting the OOC after the last written word until it finally caves in, try it.

Yep, hyperfixation is a one of the things that most LLMs are guilty of, you can easily get rid of it by using one of these two methods (or both):

• edit the message, remove the mention of the homework/historic event until the Ai let go.

• use OOC in your response, OOC is a very powerful tool that you allow you to do whatever you want in the story, here's an example:

"alright, we are done with this homework, finally we can switch to other things, yeah?"

(OOC: do not mention the homework that revolves around [historic event] ever again in the story).

I mean if it's just Llama 3, it will feature context window of 8k tokens, not a big uplift from Llama 2 and its minuscule context window, I heard it's very impressive but it's being held down by its small context window, if it's Llama 3.3, then yes, it's 128k context window size would be a game changer.

I agree, people have been waiting for over a year now since one of perchance's reddit mods said it will be updated, I think the mod's comment goes back to early/mid 2024, even if it was Llama 3, it will be a welcome change, but I feel like that Llama 3.3 will provide future proofing until the dev update to Llama 4 or 5 two or three years from now because of its superior context window size. The rapid progress of Ai is terrifying, it's like going from 8 gb of storage all the way to 1 Tb in a flash, don't be surprised if the upcoming models (not just Llama, all LLMs in general) have billions and billions of context windows, which will probably be enough to write someone's entire life story.

I used the pre-update furry artstyle because the humanoid face effect makes very decent and detailed human characters, I'm not using it to make anthropomorphic animals, I did use your method to force the image model to make humans like: (male paladin, no animals ear, no fur, no snout, no paws...etc) and ended it with "human face, not furry" but the generator bricked (still generates anthro animals).

The dev have all the time in the world, I hope he makes it right, if the new model acts, lead stories, describe environment/surroundings and animate characters differently and precisely as described (I'm talking 180° difference), then the text model update has the right to be called an "update".

(or maybe dev even put the old model back into place while working on the new one, idk).

That could be right actually, because I have been using ACC and Ai RPG since late 2023/early 2024, so I have a lot of experience on how the old model behaves, talks, describe, steer... Etc and I can confidently say that the old model is the one that we are toying with rn and the new one could be on back burner, but sometimes I feel like it's being partially deployed and then it goes away.

so far, the best experience I had with it was on day 1, when it was first implemented, it was soooo damn good, the responses took longer to generate but I didn't care because they felt so damn real, if the model behaved like that after the update is fully implemented, other ai rp and chat platforms will face a tough competition.

Yup, that what I always say in the subreddit, even before the change, the problem is that people don't listen, it's an absolute miracle that perchance exists, everything is free without the tiring censorship, so you can absolutely go ham with grimdark and mature stuff since gore and violence is considered NSFW to many LLMs, not to mention the image gen, it's all free, no subscription, no pay walls, no monthly fees, no points/tokens system and no "one purchase unlock", some people are really ungrateful. I really hope the dev doesn't get discourage by the barrage of NPCs that won't shut up about the ongoing change.

Hello VioneT, Thank you for your input! I appreciate the reply, I don't know if my post was misunderstood or I'm missing something, because these examples/suggestions weren’t requests for workaround/fixes/techniques for the current model, I'm aware of its limitations and quirks, I'm simply proposing a native support for the listed suggestions into the upcoming gen text model and inquire about the feasibility of their implementation without a hassle for the devs, as implementing them would enhance RP potential and give the user general QoL improvements, not to manually patch current model with codes, workarounds and janky micromanagement, because they make using ACC annoying, feels like a chore and ruin immersion (I already use many techniques and fixes, I've gained experience because I use it so much, I even discovered some myself, also user feedback in the subreddit and trial & errors, so far many of them are hit or miss, some doesn't work, some work when they feel like it, others work but lobotomise the bot beyond belief and others demand constant micromanagement, you can imagine how much of a chore that can be. I will be completely honest, I've never used codes in ACC because 1) I have zero experience with coding 2) codes provided by chatgpt and deepseek are ass and brick the bots that I make 3) i don't want beg and nag for codes from users, that why I made this post, to see if the suggested fixes can be implemented or not. If the next Gen update made the Ai in ACC smarter but still retain all of its issues and quirks, then the update won't do it justice. The main goal of the post is to give insights about the possibility of improving AI behaviors and changing them for the better, without user-side fixes, codes and workarounds, it's about what can be natively implemented.

Ok now I get what you're saying, I just wanted to clarify myself in case something was misunderstood.

It kinda suck that we would still need to micromanage stuff, but it is what it is, it's not all sunshine and rainbows, but in my opinion, if ACC is to move past the flawed sandbox that it is now and move toward something that eclipses AI Dungeon and other ai chatbots, it needs core overhaul for stuff like absence, tension, disconnection, and character state awareness, because treating all characters as omnipresent, immortal and clingy conjoined twin gets old really fast as you try to do something alone in the story (the narrator doesn't help).

I hope the devs will give their insights on the matter.

OOC is one of the tools that everyone in Ai rp communities knows and use, are you sure that you used it correctly? Because I did use it multiple times to fix this exact problem.

It could be deepseek, because Chloe (the ACC assistant) gets very scared and avoidant when topics like tiananmen square massacre, uyghur genocide and Taiwan get brought up, not to mention it functions similarly to chatgpt as many people have noticed, also claude and Llama 3.x/4 won't have a problem talking about the aforementioned topics, unlike deepseek. As for the token size, we still don't know, it really depends on the model and the hardware capabilities of the dev's setup, it could range from 8k-16k all the way up to 128k.

Unfortunately I don't have any images, I used to have many, brave thought it would be a good idea to fresh install itself with a new update, I looked up ACC bot share chats, hoping that someone shared a character drawn in this specific artstyle, sadly everyone uses anime or something from the new image model, the effect came standard in the painted - furry artstyle, once you click on effects, you get few effects you can choose from, like ethereal, multi-view, toons... Etc, the humanoid face is one of them, basically it changes the generated characters from anthropomorphic animals to just normal humans, it's a mix between digital drawing and oil painting, without brush and sketch marks, they looked clean, it doesn't look like anime, does not include eastern facial features that makes characters look like soft anime and most importantly, the generated characters look normal, no bimbos, no Instagram bodybuilding models and no twinks, it gave diverse faces, drew ethnicities very well and drew fantasy races like elf, dwarves, goblins, orcs...etc very cleanly, the closest thing I could find is RPG/fantasy video games art, it looks a bit of league of legends characters art (but less flashy) mashed with pathfinder games art.

These are the closest examples I could find online:

Take a little bit of the details out and make it a bit glossy, just a tiny bit, not too much, increase or decrease the saturation (depending on how gritty or light your prompt is), with all of that done, you would get a very close results to the style I'm looking for.

It looks like the dev is in the middle of implementing it :

https://perchance.org/image-gen-update-aug-8th

People saw this announcement in the AI image generator

Good luck with ST, honestly I didn't have a good experience with it (technical issues and bugs), I will probably give it a second chance, given the current state of perchance's model.

Hello Petra, I appreciate the technical clarity you presented, yes I’m fully aware of the ACC vs. model distinction, and yes, I've aware about the multiple ACC forks that exist.

The issue is that ACC is being treated like a general chat app, something like c. ai, you can make an argument like "that's how it supposed to work" but we all know how it's being used by looking at the users, we can't ignore that 90-95% of the userbase are using it for narrative driven rp, lore rich stories and looong dramas, ACC have lore entries, instructions and reminder notes that enhance the characters beyond casual conversations with your OCs, if “ACC wasn’t made for that” then that defeats the purpose for its users, the original post was about the possibility of evolving/improving ACC during or after the text model update based on how it's being used by the majority of users, hitting two birds with one stone. I've seen countless RPs, slowburn stories, and interactive novels that are littering the subreddit and being shared among users, despite the limitations and quirks. I'm aware that ACC is limited and requires knowledge in coding, babysitting and whatnot to get the most of it and change its behaviours through added functionalities and features, yes, technically, ACC was not meant for that due to its limitations and the way it is, but in my opinion, the userbase defines utility, it needs to evolve.

A better model will marginally improve ACC since it lacks a sophisticated interface and the logic to compliment it, I've tried an Ai chat that uses LLama 3.3 70B (128k tokens) out of curiosity and the character still behaves exactly like the ones in ACC, no surround awareness, no Location-based dialogue and no dynamic presence, I was really deep in some dungeon and my dwarf companion that's waiting for me miles away at the entrance keep spouting things like "don't yee forget about me! I'm waiting for yee, me bottle of ale keep me company" and shit like that, despite the fact that the bot was clearly told to stay put and don't following me and the obvious distance between me and him, he still acts like an omnipresent entity, this proves that if the new model is used within the current limitations of ACC’s interface, the change will be minimal in this specific case.

It’s great that perchance has many talented people that are patient enough to make forks, introduce solutions, figure out fixes and share codes to make ACC more tolerable and user-friendly, but if we really need to have coding literacy, constant modifying through forks, manually fix everything, heavily depend on band-aid and DIY solutions, then perchance will stagnate and rot, it won't thrive at all, which will discourage new and old users from using it and seek alternatives if it didn't evolve, there's two things that's keeping perchance afloat: full creative freedom and limitless usability, unlike a more sophisticated but limited use tools like SillyTavern, which has plugins that fixes few issues currently present in ACC, many people stopped using ACC and are waiting for the new model, we both switched to SillyTavern, but the limited uses per day, countless issues and bugs (not related to the interface or models, it hates my phone and pc for some reason) really suffocated my playthrough/sessions.

Finally I want to make something very clear: I'm not criticizing, demanding or making a wishlist for the devs to fulfill and hand them back to me in a golden platter, I'm presenting couple of issues and suggestions from my experience as a long time perchance user, I can't complain about something that kept me entertained for countless hours and I really want perchance be the pinnacle of ai chat/adventure as many apps, websites and "games" are emerging with the same premise, each one keeps innovating and making claims and add lucrative features to entice users. I made this post to see if it's possible for the presented issues to be natively fixed on ACC, without shortsighted fixes and the usage of different forks. With all of that said, I'm really grateful for the help and tips from you and VioneT, I greatly appreciate that, the only thing we can do now is to wait and see how ACC deals with the new model.

Aye, glad to help

90% to 95% of the user base actually uses ACC for NSFW and sexual stuff with their characters lmao

Well it looks like I underestimated perchance's goon squad, I'm not in perchance's discord neither do I visit nsfw tabs in ACC, but I do see many users share their OCs and real life video game and anime characters, some even share the lore/characters in the subreddit, the only nsfw stuff I see are posts on the subreddit, where users ask questions about the image Gen model and how to make celebrities, but you are right, Ai RPG does need some TLC, it looks like some obscure text app from 90s, if it received the changes it deserves then I feel like it will compete/compliment ACC.

The vibes resembles the artstyle, especially if you choose to generate fantasy characters where everything is a bit saturated and glossy, minus the anime face of course, I did post few examples to Danno down below, it's the best I could provide.

Tried to generate a couple, none of them came close unfortunately, on the bright side, I was able to snatch a few of them from the subreddit, none of them had the humanoid face effect applied but its better than nothing, here :

The generated faces didn't have a default face like the new Gen, it gave a diverse set of faces, none of them looked the same, now take the faces from these images :

And mash them into the anthropomorphic sailors/captains images, that exactly how it looked like.