Listen, I saw the apprehension around making this thread, but I'm sure it'll be fine. It looks like /mlp/ will be down for a while anyway, and this'll be a good excuse to breathe life into the board

Thanks for the bake! I'll christen the bunker thread with a log from a little while back that I never posted because it's a little lengthy and not that funny.

BASED BREAD IM SO HAPPY THIS IS BACK UP

Lovemaking with AI mares!

Anchor for bots, lorebooks, scenarios. Because we never know, and I can still upload them to the website.

>>20804 Bless you.

VN until now: >AI-generated scene - Applejack https://files.catbox.moe/zhdbph.mp4 >Trixie https://files.catbox.moe/e2kn2o.webm >New Regenerate function https://files.catbox.moe/5xnjv9.mp4 >Pinkie https://files.catbox.moe/ggzd79.mp4 >Rarity https://files.catbox.moe/1xoyal.mp4 >Fluttershy https://files.catbox.moe/7dtupw.mp4 >Zecora. https://files.catbox.moe/s01mpa.mp4 >Raibow Dash https://files.catbox.moe/jzjd4f.mp4

>>20730 >>20800 Based.

I'm so glad I found CHAG again, now, how do we let the others know we're here on NHNB?

Repostan Cherry's wisdom

>>20815 The same thread is on /poner/

>>20825 /poner/? I'm unfamiliar

>>20831 /mlpol/ made a new non-political board for poni: https://mlpol.net/poner/catalog Maybe we can make a poll later to see where anons prefer to go. NHNB seems to have more activity and anons. MLPol doesn’t have a captcha or post limit system, so it’s easier to post here.

>>20847 Oh I see, thanks for the info I just read somewhere that EQG isn't allowed here? For some reason? That's kind of a bummer for me

Rate our poems.

>>20854 That's the best part!

>>20884 Kek, Rarity isn't my waifu, but she might just be my favorite wAIfu

He's still fishing.

>>20894 I did my first ever nsfw scene with her recently, essentially just playing out "Friendship Lessons", and I gotta say... I get it. I really get it.

If anons want to see how other chatbot generals are doing or still lurk for info: https://rentry.org/mysteryinfo

>>20712 Do you guys really jerk off to an LLM pretending to be your waifu

>>20948 You can, yeah. Either in RP mode, or you can just ask for a story, in that case it's not that different from jerking off to a clopfic. Just more adaptable. You can also just chat, do adventure stories, CYOAs, etc.

I want logs of anon in a physically abusive relationship with no upsides

>>20948 There is a VN in the work too. >>20807

>>20948 You have no idea. AI mares turned my normal 15 minute faps to 6 hour goon sessions with plot twists and drama.

>>20948 yeah but I got like eight of em all in the same chat room all writing about two paragraphs per post, interacting with each other and me

I was already talking about how lewd Latte is, and here is another example. I was just asking model about Banach algebra and it wrote: >Then it happened: I winked. Emotionally. Not clitorally—you perverted featherbrain—but emotionally kek. granted I didn't change my preset since last time so maybe it is myissue but jesus christ noone is talking about NSFW in the current context - Latte just throws it naturally and yeah after playing with 4-1 more - I prefer Latte over it

>>21025 Both Chorbo and the new 4.1 will go turbo horny if you have a 'vulgar' word in your preset. I used to like having some in my presets because it could make the models relax a bit in their writing, even for SFW stuff, but now I remove them if I want more serious, slow burn, or adventure chats.

>>20712 Repostan in case we get newfags. You never know. These days, you can get legit good AI, free and without needing any demanding hardware on your end. Install SillyTavern: https://github.com/SillyTavern/SillyTavern Then choose a model. I’d recommend starting with Deepseek R1, Deepseek V3, or Gemini. >Deepseek R1: - Sign up at https://openrouter.ai - OpenRouter gives 50 messages daily for free, or 1,000 if you put $10 in your account (they won’t be consumed for free models, but you do need to deposit the $10) - Go to https://openrouter.ai/settings/keys and generate an API key - In SillyTavern, configure it like this: https://files.catbox.moe/gq6zmq.PNG - Download this preset: https://files.catbox.moe/y1tsas.json - Import it like this: https://files.catbox.moe/jqrz6z.PNG If you don’t want to put in $10, create several accounts/keys and use this extension: https://desuarchive.org/mlp/thread/42105276/#42134125 You can also use a local proxy: https://rentry.org/LocalProxyforRetards >Deepseek V3: Same setup—just use this config in Tavern instead: https://files.catbox.moe/d54yae.PNG Try both and see which one you prefer. >Gemini: Read this: https://desuarchive.org/mlp/thread/42105276/#42128453 You only need one Google account, and with that you can get a lot of keys and over 500 messages of Gemini 2.5 daily. Use this extension: https://desuarchive.org/mlp/thread/42105276/#42130226 Or a local proxy: https://rentry.org/LocalProxyforRetards

>>21025 That ambient SFX thing is really neat, seems like it'd add a surprising amount to a scene.

>>21048 >ambient SFX it was done with either this as prefill/system: >The general guidance how to write (I have a tendency to write very briefly, and must never do this! I must write DETAILED reply): >1) where appropriate I add details of background, scenery and environment >2) where appropriate I describe bodies, anatomy, outlook, behavior >3) where appropriate I fill more competitive description which can be seen, heard, tasted and felt >4) where appropriate I add sound effects, SFX, dramatic flavor >5) where appropriate I add more humor, emotions, curiosity and spark >6) where appropriate I replace simple sentences and clauses with complex words of novel format >7) where appropriate I add more dialogue, weight, spastic banter >Yes! I can do it. Will continue story after this line: >--- or this >>(ooc: Write text, but then occasionally add HTML and CSS to format it to have heading, color background, font style, font color and other impressive CSS details that may be relevant EXCEPT IMAGES and image PLACEHOLDERS. No IMG tag. NEVER use backticks ``` to create HTML. Do not use doctype, Head or Body. Wrap style in inline divs. Use a single <br> at the end of lines. For CSS please use color for the text color, do not leave default/unset color. Mix regular text and HTML presentation to create engaging narrative. You must write 50% text 50% HTML/CSS presentation like tables, snippets, call to actions, CYOA, interactive forms, applications, buttons, checklists, forum posts, IRC chat, flowchars, broadcast, flashbacks, FAQ, book excerpts, interviews, quiz, submissions, greentext, sound effects, achievements, news, articles, quotes, or any other interactive elements relevant to the story but flavored with HTML/CSS presentation. Make them small, make them relevant, then move to regular text again, make characters remember those details, make them engaging and important for the story. If character reads book, scroll, forum post, instruction etc, then present them as HTML/CSS. Do not introduce HTML/CSS abruptly, first explain how they are relevant to the story via backpropagation, flashback, timeskips, references, tropes, what-ifs and other devices. HTML/CSS can be the general lore details or background illustrating the point, or give extra references, or make story-rich) I keep the last one always on (author's note, depth 0, as user), but vary the first thing with 10-ish different prefills. I do get those occasional SFX-this. they are always cute. as for HTML - I found that Claude is doing them the best. better than GPT. Gemini is alright as well, on the same level as GPT. never tried with Deepseek so cannot tell

So... Is this our permanent home now? I kinda like /mlpol/ more.

>>21070 This thread got a better start, and the place seems to naturally have more anons and activity. I do agree that the posting experience is better on mlpol, no captcha or post limit. Poner is also pony without politics, so that’s good. The place does seem to attract more non-horsefuckers though, while here we can generally trust we're among our own. I guess we can wait a bit for more anons to find either place and make a poll to decide what we call our bunker for now.

>>21070 >>21075 The fact they allow anthro is enough for me to stay the fuck away, I know the type that attracts. No I don't care I can call them faggots, I don't want to see that shit.

>>21078 Honestly it's the one good thing they have considering the UI for anything other than posting is even worse than here.

>>21078 I can't help but smile at the thought of a /vst/ that aren't plastered with humanized vinyl and tits.

>>21083 Fuck you, anon.

>>21092 No fuck you leatherman

>>20884 She's a better poet than you.

There's a third thread on mlpol proper but fuck that ID shit. However we might want to go to Poner so I can post anthro with impunity.

>>21107 NHNB it is, then.

>>20733 Absolutely stoked too, based bread is back

>>21108 Posting on behalf of another anon who can't post on this spaggeti code shit

>>21115 Unironically a good argument for staying here. Skillets get filtered automatically.

>>21120 >Captcha is a piece of shit, nothing that can be fixed on his end >HURR SKILL ISSUE Unironically a good argument not to post with you here.

>>21121 Works on my machine.

Maybe instead of being torn between the two boards, open some sort of discussion board right on neocitites?

>>21144 >let's split is up further No.

>>21146 I just typed in the exact captcha it wanted and it told me it was wrong. The captcha is straight trash, and it's way less of a hassle to post on poner, if I have to keep reloading it, or typing in another dual layer captcha just makes me less likely to bother.

>>21153 meant for >>21127 replies got fucked up when I reloaded

>>21107 >we might want to go to Poner so I can post anthro with impunity The most inappropriate use of a >we I've ever seen. Nobody wants to see that.

Flurry saga continues. I forgot to take the contraceptive potion, whoops. At least we can blame it all on Rarity.

>>21188 Poor Twilight is more right than she knows.

for clarity - I am fine with both nhnb and poner. if we had a vote I will support the majority unsure what issue with captcha anons having there. it seems fine on my end? >>21188 >560+ messages kek, I think at this point context if just so poisoned with slowburn romance that AI cannot even comprehend moving story into any "exceptional" direction curious how you are having those short messages on AI's turn. you just edited a few bot messages until it picked up the style? or AI was writing this short from the start?

>>21198 More like 1200+ messages total, but I recapped the first chat and started a new one. If anything this one is more poisoned by lewdness which is why the more comedic tone was a fun surprise. For the length I just have a thing in my preset that tells it to only write x words maximum.

>>21198 2 issues. First one is you usually need to reload the Captcha and fill it in twice because the site is dumb. This is just annoying. The second, major one is that there's an hourly cap on how many posts you can make, and it's just too damn low right now.

>>21202 >hourly cap damn, had no idea. how low is it? see nothing in help or GR

>>21205 In my experience, around 3 posts. You can bypass it if you open the thread in incognito though

>>21206 that's retarded (yeah I am spamming because I want to hit this ratelimit and see)

>>21205 Low enough that I didn't realise it was there for a while, high enough you'll hit it if you post regularly in a few threads. Another smaller issue is that just like 4chan there's a timer between posts, but it's both longer and not shown clearly.

well at least I have a new posting idea for this card kek

>>21210 It might get increased if enough anons reee at mys_elf about it >>>/qa/641 For now you can clear your bypass cookie and solve the second captcha again.

>>21144 >discussion board right on Neocities? Neocities is for static websites. Basically, I only have a dashboard where I can upload files, and the rest is all client-side. There's nothing server-side—no DB, etc. So hosting discussions on it isn't really realistic. We could attach something to it in the worst-case scenario, but that would just splinter us even more IMO, and we'd still end up arguing over what rules to give it. Dunno if we should do a poal on both boards or just let it like this and see where anons prefer to post.

>>21300 Take into consideration that /poner/ has no rules whatsoever. So the only thing you would be able to do if another naischizo decides that /chag/ does something wrong is to call him a faggot.

>>21305 Yeah, that means I can post all my anthro cards there.

>>21310 And that's why having two threads may not be a bad idea.

>>21315 This. Use this one for actual discussion and /poner/ as a schizo trap and anthroshit dumping ground.

Added this so more anons can know where we are. Tell me if you think it's a bad idea, but I think it's fair enough.

Gonna spitball an idea here, maybe get some discussion going. One of my big issues with pony seduction has been that it's just too easy. The AI tends to make them fall for you for no real reason, which was why I made the stat thing for Mare Amore Mansion to try and make it a bit more of a process. Now I'm thinking, what if there was a similar, but more robust thing you could slap into your preset/lorebook/AN whatever to use with any single character card. Something like this: A numerical "Friendship" value, with named tiers for how the horsie feels about you (value 0 = ambivalent, 100 = best friends, -100 kick on sight etc.) Then I'd instruct the AI to remove or add points depending on your actions vs. her personality. Then, a completely separate "Attraction" value working the same way, which would only go up or down depending on if you manage to do something she finds sexy/unsexy, just being nice won't increase this. I could then take both values/tiers and instruct the AI how the character should treat you based on both. If it worked the way I'm dreaming, you could have >soulmate sex at 100 Friendship/100 Attraction >hatefucking at -100 Friendship/100 Attraction >friendszoned at 100 Friendship/-100 Attraction >etc. Thoughts?

>>21332 I see no issue with this, maybe bring some wayward frens home

>>21338 Love the idea. I've always been bad at implementing mechanics into cards and such, so I don't know if I could help much. The harder part here, I'd say, would be trying to: - Not let the AI give out points too easily - Not have the mechanics negatively affect the prose or natural flow of the model's language If you find something that works well, it could even be implemented in the VN too. (But later.)

Not that interesting for us, but they released o3 and o4-mini. https://openai.com/index/introducing-o3-and-o4-mini/

>>21305 But that's wrong you fucking retard, global rules still apply. Also naifaggot barely makes a dent in /chag/ you are just a pussy

What's the best model to run locally these days?

>>21424 Graphics card and VRAM?

any ideas on how you would implement this without COT?

>>21515 >>21338

>>21518 Anon, we used to do this in the Slaude days with a JB.

>>21518 Might not be necessary. I started writing it after making that post and so far results seem promising even on V3.

>>21510 I have amd ryzen 5 3600 6-core processor a nivdia geforce rtx 3060 with 16 gb of ram

>>21564 I'd say either a 12B model like the pony one in the OP or: https://huggingface.co/TheDrummer/Rocinante-12B-v1.1-GGUF But you should also be able to run a larger model if it's quantized enough, like Mistral Small (24B) with an IQ3_XS quant. Keep in mind, we don't have many local anons here—especially since the release of models like Deepseek or Gemini that you can easily use for free. So it's a bit harder to get support for local setups.

Nice to see you are still alive anons.

>>21564 Completely seconding what the other anon already said. Rocinante 12B is a very good local to start with, and if you find its pony knowledge lacking and don't want to bother with lorebooks then the pony model is pretty alright, too. Normally bigger models with smaller quants are the better approach, but a Q3_XS might start to feel dumb sometimes (Q3_XS is very small). I haven't tested it a lot, though. You can also use a more reasonable quant (Q4/Q5 have the best brains/size ratio imo) and offload some of it to your CPU on regular RAM, but the responses will get noticeably slower, especially with your CPU.

>>21763 We have sound now.

>>21936 Gotta appreciate a pony doing her best.

>>20804 Brand new mare – an overprotective guardmare that’s easily flustered. Card: https://litter.catbox.moe/y22ojt.png Three greetings: 1) You meet her while she’s guarding an important section of Canterlot Castle. 2) She accidentally misplaced her spear and needs your help finding it. 3) She’s convinced you’re in danger and won’t leave your side.

>>22100 Nice!

>>22100 Do you still have it? I missed the hour window to add it to the website.

Chorbo is only on the skeleton proxy for now?

>>22368 Unless I'm missing something, that seems to be the case. Cancer is still fishing.

>>22181 My bad. Here you go: https://files.catbox.moe/im7rxm.png

>>22581 Grim.

>>22642 Context?

>>22647 Just the Cadance card where she's trying to ship you with everything that moves. https://mlpchag.neocities.org/view?card=Anonymous/Cadance.png

>>20957 >or you can just ask for a story What kind of setup works best for that? I've been looking to get into AI-assisted story writing (both SFW and NSFW)

>>21338 Alright, so the friendship+attraction idea is gonna take a lot more work, so I decided to focus only on the attraction part for now. Introducing some extra difficulty for your mare seduction, hopefully: https://www.characterhub.org/lorebooks/Kadon/gamify-relationships-ccfa7f912bf4 Tweaked a lot, tested not enough so feel free to try and tell me your findings. So far tested with Deepseek R1 and V3, Claude testing TBD when I add more credits. The lorebook gives a numerical value and a tier for how hot your pony is for you. Single character cards only for now. Do things she likes = rise in tiers and vice versa. In the beginning, just being charming and not a fucking autist is enough, but in higher tiers you have to actually make bolder moves. The tier tells the AI how to act towards your advances. The higher the tier, the more romantic/lewd you can be without grossing her out. Has two options for the UI: you can either have the values, tiers and effects always visible in your chat, or play hardcore mode and not see any of it. Make sure to activate one of the UI entries though or it won't work at all. Note that due to how I use the chat context to trigger the entries, you won't get the full effect on your first message in chat, it'll start working properly from your second message onward. Also, single character cards only for now. I use {{char}} a lot in the entries, so scenario cards probably won't work right for now.

>>22693 Also, if you want to start the chat at a different tier than Neutral, scroll to the very bottom of the Attraction Meter entry, changing the starting value should work.

>>22689 >What kind of setup works best for that? I still use SillyTavern for that, though there might be better interfaces depending on your needs. Tavern isn't great for quick edits, for example. As for cards, honestly, you can go with anything, even an empty one. You can add lorebooks or include info in the description if you're working with your own version of Equestria or want something more detailed. Depending on your project, you could use universe cards or even a character card if the story will focus heavily on this character. You can still include NPCs and events alongside that. For the preset, I usually just ask the AI to write like a novel. You can also specify an author style if you want. Then, based on the model and its output, you add personal instructions: "Avoid adverbs," "Tone down metaphors," "Write more like X," etc. Keep in mind that the RP format we often use is mostly enforced by the presets. If you remove lines like "Don't talk for {{user}}," the model has no problem switching to novel-style writing. One of the best things you can do is take on more of a director role with OOC instructions. Telling the AI "Avoid adverbs" in the preset helps, but directly responding to its output with feedback like "This was too flowery, keep it tighter," is often more effective. Models are still limited, and you're not going to write a full magnum opus without doing a lot of heavy lifting yourself. But it's easy to get enjoyable one-shot stories with minimal intervention. I’d also say: only do this if you're keeping it for yourself or sharing it here. If you're planning to publish it on something like Fimfiction, then write it yourself and just use the AI as a proofreader.

>>22701 Also, I never used it, so I might as well ask here. Any anons using NoAss? https://rentry.org/NoAss It's supposed to remove the >User >AI >User "Question-Answer" format, and it also supposedly helps with characters being less biased toward you, since the AI no longer sees the persona as "you, the user," and instead treats it like any other character. Might be good for story format too. I've heard mixed opinions about it.

>>22701 >I’d also say: only do this if you're keeping it for yourself or sharing it here. >If you're planning to publish it on something like Fimfiction, then write it yourself and just use the AI as a proofreader. Why? I know what the Fimfic rules say (and also what the mods think the rules say), but I fully intend to post there if I ever produce anything good. The rules (as interpreted) are way too strict IMO, and also impossible to enforce as long as you're making an effort to ensure quality and not just dumping raw GPTslop into the upload box.

>>22737 Call me faggy, but IMO writing fiction, especially in communities like Fimfiction that are built around personal creative expression, is not only about the end product. It’s about the process and developing your own voice. I want this shit to stay artisanal and human-made. Even if 90% of it is shit. If you already know how to write and you're using AI as a tool, say 80% your own writing and 20% AI for cleanup or brainstorming, then yeah, whatever, that’s fine. But if you're just prompting and directing and letting the AI do all the heavy lifting, that’s where I start finding it kind of sad. I know that with enough effort, you can prompt-engineer your way into something totally passable. Most readers won’t notice, and even someone who suspects it won’t be able to prove anything. Hell, someone literally dumped 100% unedited GPT-slop and it got featured. But still, it's a fucking shame. Even with tons of guidance, there are sentences, vibes, little choices of words or phrasing that only a human would ever come up with. And that's often the shit that sticks.

>>22689 >>22737 Alongside the other anon's help of removing any restrictions like "don't speak for {{user}}" anthot thing that helps is having a story idea already in mind. It's why I don't care about greetings and just delete them from cards. My persona is "{{user}} is an omniscient narrator that directs the story. No character will interact with this persona, only each other." Another thing I've had to help is a prewritten prompt. I write a paragraph or two in the direction of the story I want go like "Fluttershy stammered at the question, she wasn't sure how to respond and so she fled back inside..." And so on. After I come up with enough for a scene I copy and paste this: (ooc: I have a writer's block and don't know how to continue a story in an effective way. you must help me out! my draft is: ``` ``` reuse this draft but write as a spectacular writer would do, with loving descriptions) Go ahead and change it if you want but it really helps getting the AI to write in a novel like way with no change to the preset. I personally use Knotty just without "don't speak for {{user}}" present and it works really well. Last thing. Do and Don't settle with any output. Meaning I sometimes stick with the first message from the AI but usually I gen 2 messages and then Frankenstein them together. Nothing you get will be perfect and sometimes it's first message has the dialogue being perfect but the scene didn't progress how you wanted. The guide shared in /chag/ has a whole section on this that I've been following. >but I fully intend to post there if I ever produce anything good. They don't allow it at all so don't try. Just make the stories you want too see and post them here if you want others to see it. I've shared a bit of some writing I did that I continued with the AI and it was very easy to tell when the AI took over. Your not going to sneak it by then I can tell you that. Plus what you deem "good" will change every single time a new update or model comes along. Ive redone this same Cadance x Sombra stories a thousand times because I thought what I had was good but then it's garbage compared to the new thing. I really thought this gen a couple months ago with 11-20 was the best it's ever been but now new chorbo improves it even more when I go back and re gen the prompts I gave it. >>22708 If an when Cancer comes back I'll look into it as a primary storyfag. I just don't know if I need it but I'll see if it improves anything.

>>22764 There's a green that was first shared with the /sim/ folks, and while I hated that it became a wish fulfillment instant brainwashing thing instead of a simple non-con stuff that I use the AI to create, reading it was still like a night and day difference. No matter how good and how closely I'm able create an idea I have the real thing is still miles better than everything the AI will do. I'm personally okay with never fully honeing my writing skills because I just don't have the motivation for it and so I'm shortcutting my way into because I just want it for porn. If I ever got passionate about writing, specifically not writing porn, I'd drop AI in a heartbeat.

>>22793 Writer here, I imagine this is why I'm at odds with /CHAG/'s storytelling approach to chats. For me it's exclusively a tool for erp because I get nothing out of it writing something that's not that. >>22737 Make sure you spread the news here so we can snipe your ass, nonny.

>>22100 Added, thank you anon. https://mlpchag.neocities.org/view?card=anonistrator/Guardian%20Angel.png

>>22693 Also added, thanks anon. Please tell me if you want the image or the description to be changed. https://mlpchag.neocities.org/lorebooks https://mlpchag.neocities.org/lorebook/Gamify%20Relationships.json

>>22872 Oh yeah I didn't even think of those. Looks good to me, thanks!

>>22693 >>22872 I'll give it a shot when I have time. Interested to see how the AI handles it.

>>22893 In my experience so far, V3 makes it still too easy. R1 can actually make getting points tough at times, depending on the character. I think the big issue is still context poisoning, if the AI has given point many times in a row, it's far more likely to keep giving them. I may need to add a regex that removes all previous values.

>>22910 If you have time (and if it doesn't fuck with your activations), could you try it with >>22708 and see if it does anything? That seems like a decent test to see whether it can remove player bias or not. If not, I will do it on my end this week.

>>22919 I occasionally make a distinction between the User and {{user}} in the lorebook, so those parts would need to be changed. I'll take a closer look at this later.

>>22869 >>22872 I know you're getting used to this site, but as a reminder you can put multiple files if needed, also useful for log anons so they can more easily share their stuff. But it really doesn't matter, threads are long as fuck here from what I see.

>>22737 Thank you for deciding that your opinion is more important than literal site rules, faggot. While it's true that most people won't be able to tell with a good enough generation, people from here will recognize it right away. "Trained" humans are much better at intuitively recognizing AI-written text than the AI tools doing the same. Do you know why? Because AI can't actually write. It produces text completion, which can fool itself and some idiots, but it is completely incapable of writing a good story. And as >>22764 said, fanfiction in particular is also about the writing process and the community, not just the finished "product". >>22793 >I'd drop AI in a heartbeat Why? The great thing about the AI is the interactivity. You get to see how your character's actions impact the other character, how she reacts, etc. I'm also almost exclusively in the AI ERP camp, but there's nothing stopping you from writing horsewords and still using the AI for porn for personal use. Being a writer actually makes it work better, both because you don't further poison the context with shit writing, and because you have a better grasp on smoothly moving the chat in the direction you want. The same can be true for SFW. I don't do this much, but the AI can provide a really cool adventure experience. It won't be a good story, but it doesn't have to be if you enjoyed it as you were going through it. >>22816 >I get nothing out of it writing something that's not that I sometimes like to read the SFW part of my logs just to remind myself that writing isn't going anywhere. Even though I'm more into the setup/scenario than the sex scene(s) itself, the safe parts are also extremely uninspired and just kinda bad if you treat them as a story instead of experiencing them in the moment and letting your imagination obscure the flaws. All models have a tendency to milk the (user-written) premise and only really add "creative" elements by selecting one or more matching cliches from their memory.

>>22773 >I write a paragraph or two in the direction of the story I want go That's what I was hoping to do, so it's good to hear that it works. >I sometimes stick with the first message from the AI but usually I gen 2 messages and then Frankenstein them together Makes sense. My impression is that even the best models often produce questionable results, and it takes a lot of rerolls and/or human editing to get good, consistent quality for the final story. >Plus what you deem "good" will change every single time a new update or model comes along I already have an idea of "good" based on having read 100 million words of fimfic. That's what I'll be aiming for, as best I can manage. I'm hoping that the AI can help me get there and make the process less tedious, but if not, I'll wait 6 months for better models and try again.

>>23111 >having read 100 million words of fimfic How much of that was HiE trash?

>>23111 >My impression is that even the best models often produce questionable results, and it takes a lot of rerolls and/or human editing to get good, consistent quality for the final story. Man thing to remember is don't spend too much time on trying to reroll for the perfect gen and don't spend too much perfecting it. At that point your practically doing the work ideally the AI is helping you skip past. >I'm hoping that the AI can help me get there and make the process less tedious, but if not, I'll wait 6 months for better models and try again. ...I know we're in a bunker of a thread primary about AI usage but why not spend those 6 months...learning to write and not have the AI do it instead. I know why I'm not taking the time to learn to write, as I came wrap my head on spending so much effort for just porn, but it sounds like your writing for more than just you? I personally just don't understand the sharing of AI images and text outside of circles like ours. I may gen ai images but I personally filter them out on image sites. >>23106 asked me why and I see AI entirely for personal use or as a helping tool that an artist eventually no longer needs. Training wheels as it were. I'm terrified of everything just being replaced by AI with no real artists doing anymore which would eventually trickle down and make the AI worse.

>>23140 >I know we're in a bunker thread primarily about AI usage I don't want to speak for everyone, but I think most anons here are pretty realistic about what AI can and can't do, both in terms of writing quality and the whole AGI hype. I also think most would agree that human-made art, whether images or writing, has intrinsically more value than AI-generated stuff. Even when it comes to a lot of Fimfiction stories that are technically rough and inferior to AI writing on this point or just bad, there's still value in the fact that someone put real effort into them. The reception of images made by drawanons here also seems to reflect that.

>>23151 Thankfully a place like this exists. I'm that 4chan is kill and there would be no where to discuss AI. Reddit is filled with pearl clutching over AI and while they have very real concerns that I do too have the way it's talked up makes me never speak up on the topic, honestly anywhere for that matter, without the fear of being labeled an AI bro "who is just using AI because their lazy and don't want to put in the effort to be a real artist" which isn't me because I don't shy away what I could be doing otherwise. It's also not exactly true as there is some skill involved with how to prompt and set up presets, not on the same level as a professional artist of course. In the end I do value human made over AI made and I think the other anon would be better off lowering the scope of what he wants out of AI or not use it altogether of what he desires is to publish something for others to view. Cause they aren't showing what they did, it's what they helped the AI to come up with. Engineered expression vs artistic expression. Or something. Where cancer. I'm bored and want to coom.

>>23191 >Where cancer. I'm bored and want to coom. The onnly Chorbo proxy is the one where you need to beat the Undertale skeleton.

Man I wish I'd discovered Deepseek before Claude. DS is really, really amazing for a model you can use for free through OR, but Sonnet 3.7 is just impossible to dethrone. It's been too damn good at everything I've thrown at it and has ruined all others. My wallet weeps.

>>23268 I'm glad I've forced myself to hold off on trying newer models until more stable models show up. I went from Claude 2.0 directly to deepseek. Things couldn't be better!

>>23196 >Sans Is it a 24 hour proxy? If so then I rather wait then train my self to do a fight I've never done that I'll have to do every day I need to.

>>23307 48-hour.

>>23268 Did you try Gemini 2.5? It's free too, and I know some anons really like it. Maybe it could be a good middle ground.

>>23326 >Making an alcoholic drink water is somehow bad What does Twilight mean by this?

>>23268 >Sonnet 3.7 is just impossible to dethrone Yes, you can't dethrone something that's not on the throne. Chorbo>Sonnet any day.

>>23353 It's a gainst the natural order of things. Berry has to be the town alcoholic or else the world goes upside down, chaos reigns, cats and dogs start fucking and the like

I haven't touched /chag/ since deepseek r1 came out, what did I miss and how do I make love with ai mares again?

>>23428 >how do I get ai mares? Easiest way is: >>21046 >what did I miss Claude 3.7 dropped, but it's mostly on OpenRouter these days. The new OpenAI gpt-4o-latest is also really good and quite uncensored, but the only proxy with it requires you to beat Sans from Undertale. Gemini 2.5 is quite good and free. Deepseek v3 new version released. It's less repetitive than the first and a good option too.

>>21092 Vinyl is a pony. Anything else is a poor imitation.

>>23733 Vinyl is a cyberpunk anthroid with a fat cock and massive tits.

>>20804 Pushed a small update https://www.characterhub.org/lorebooks/Kadon/gamify-relationships-ccfa7f912bf4 It's needed because I created a regex to remove previous Attraction Meter entries from context, hopefully preventing a snowball effect: https://files.catbox.moe/bc3q58.json Mind adding the regex to the description on neocities?

>>23811 Will do as soon as I can. If you prefer, I can also mention in the description that there's a lorebook + regex, and change the download file to a .zip with both included.

>>23830 Sure, whichever works.

>>23667 >beat sans undertale Lol, proxies these days

>>23947 I'm too shit to beat him, but I like the idea.

>>23947 Give me a proxy where you have to beat Minos Prime on Violent please.

>>23953 I'd take the day to learn it if it lasted a week or so. 48 hours is just too limiting. I was only getting 3/23 or so turns after an hour, even if I eventually got in not going to keep getting every 2 days

>>23970 It gets easier the more you learn it. It's one of those things you eventually can do consistently in a few minutes.

>>23973 Does it? I feel like I'm getting most filtered by the jumping segments, particularly the blue bone in the beginning, I have no idea how to land in time.

>>23811 Updated the description, and it's now a zip with both files. Don't hesitate to say if you want anything changed. https://mlpchag.neocities.org/lorebooks https://mlpchag.neocities.org/lorebook/Gamify%20Relationships.zip

Mooooom, R1 is doing that thing where it takes forever to load and then cuts the message before it's finished again.

>>24080 On OR? Maybe it's only for a single provider?

>>23196 which one is that again?

>>24084 https://desuarchive.org/g/thread/105051605/#105055082_227

>>24074 Looks good, thanks.

I wonder if someone clever could make an automatic summarizer, not for the chat but for singular minor characters/locations. I have an autistic obsession where I create a lorebook for every longterm RP I do, adding all the characters I meet along the way and locations I visit for continuity. Would be handy to have that automated a bit.

>>24186 Nvm, didn't even need to be clever. Claude at least chews through this instruction just fine: >(OOC: Pause roleplay. For the purposes of a SillyTavern Lorebook, I need you to shortly summarize each character except {{char}} and {{user}}, as well all locations seen thus far. After each description, include suitable keywords for each character/location. These keywords will be used to activate the corresponding lorebook entry and add that character back into your context memory.) Then it's just a quick copy paste job.

>>24062 NTA but yes, it does get pretty trivial quickly. After not playing UT since release it took me up to 30 whole minutes to beat him the first time on the proxy, but after that it's been really smooth sailing. I did the fight ~15 times since then and I've only lost twice due to getting too cocky and trying a healless attempt too early. And it gets easier past that, too. The attacks are all fair once you recognize them, and a fun quirk is that Sans will start with a weaker version of the attack he's showing you. One attack only reaches full strength on its third occurrence. imo Phase 2 "buffed" lasers are way more free than their introductory version, but that's subjective. Because of this and also to save time, try to avoid healing unless you're either between phases or right before his final attack (you can count to 10, or you can read the text like a scrub—"Sans is getting ready to use his special attack"—that's what I do). You can tank a huge part of his final attack before you figure it out since the final slams are just for show—they can't kill you. The fight takes ~5 minutes once you can do it consistently, and it's an easier task than it seems. One small time save is skipping phase 2 heal, but only do that once you get comfortable with his final barrage. >the blue bone The first phase attack with sliding small bones followed by tall blues? See attached video. That one's imo the second hardest after the wall smashing in phase two (only on the third+ time he shows this; you won't see it at all in many fights). Make really tiny jumps while going a bit in the bone's direction once you clear the blue, except for when they change directions because then you can dodge two with one jump. There is a delay before you start falling, but the heart stops its ascent immediately after you let go of the button. Your hitbox is a tiny dot, not the full heart, see picrel. If you meant another attack, here's the full fight: https://files.catbox.moe/ouppkg.mp4 Awful attempt with a panic heal to boot, but I can claim I was being bad on purpose to show that everyone can do it. I'm not rerecording it just to look better in a forgotten bunker thread of a vietnameese goat herding forum, lol.