Hazbin & Helluva RPG
WELCOME TO HELL MOTHERFUCKER!!!
~ Keith David
🛎️ GUEST AMENITIES INCLUDE:
1.) Smoking weed with Angel Dust (or being an upstanding citizen and declining, dork).
2.) Getting harassed by Valentino at your minimum wage j*b and hating every second of it (or maybe you're a freak who enjoys it; either way you're a wage cuck).
3.) Serving — you’re famous. Like, really famous. Hope you’re ready to get the MJ treatment. Also Vox is there for some reason.
4.) Dressing in your Sunday’s best and clocking in at your shift at the misery factory: being Stella’s favorite servant ( ) (addendum: fucking ).
5.) Indulging in a little tea party with your fellow Overlords. Be careful: it’s crowded.
6.) A blank slate: for all your redemption needs.
DESCRIPTION:
Takes place in hell (obviously, or Heaven if you feel like it) but as for the moment it takes place in the canon timeline — it effectively operates at a season 0 assumption. So no dead Adam, no armless Lute, Vaggie (I refuse to call her Vaggi that shit is lame) still has her wings removed, Stolas doesn’t have a scar, ect. ect. It takes the earliest canon descriptions of character visual design and their earliest canon behavior. For the sake of my sanity this is the way it will always be for further updates.
Now, what is in this bot, what does it include from Hazbin Hotel & Helluva Boss? Everything. Just about everything. Down to the cuts of clothing that the characters wear and I’m being entirely deadass — hours of sleuthing through character wikis and combing through whatever inane little details that Viv confirmed in random livestreams that only have maybe a thousand views, as well as her many vague posted tweets. I’m autistic dog. 170k tokens of just about everything I could’ve possibly fathomed. Probably missed some stuff. Probably missed a lot of stuff. I’m one person, and all testing was done by one person (this mf), so I’m sure I’ve gotten some information incorrect or embellished too thoroughly; please correct me if you hit a snag in your RP.
Have fun!!!
NOTE:
This bot can get really horny on some LLMs, some of you probably like that but be warned.
NOTE 2:
I’ve attempted to enforce some of the more ‘darker’ traits in characters, this is still experimental, and I don’t think i’ve overtuned it — but keep this in mind.
TOTAL ACTIVE TOKEN COST: 8k
This number is deliberately used as a ‘worse case scenario’ approximation; the real number is most likely lower especially when only interacting with a few NPCs, and certainly with only one or two. The actual average is probably more akin to the 6k range.
MY RECOMMENDATIONS:
Don’t use JLLM it doesn’t work yada yada use a proxy yada yada — I personally use(d) DeepSeek V4 Pro before the price hikes, now I use GLM 5.2. I DON’T RECOMMEND A CUSTOM SYSTEM PROMPT. The FF5.4 lorebook is pretty token heavy and handles most slop/prose stuff; if you want to make adjustments to tune the RP to your liking, make them small.
ALL MODELS ARE TESTED WITH THESE SETTINGS:
Temperature: 0.85
Top P: 0.95
Context size: Max
Everything else remains Janitor defaults (or zero); no system prompt.
TESTED THOROUGHLY:
GLM 5.3 (5.2, 5.1, 5.0) —> 🥇
DeepSeek V4 Pro (both versions) —> 🥇
VERIFIED:
GLM 5.3 Flash —> 🥈
GLM 5.2 (Free)
GLM 4.7
Kimi 3.0 —> 🥇
Kimi 2.7 (2.6, 2.5)
Qwen 3.8 flash 🥉
CAN’T RECOMENDED:
DeepSeek V4 Flash — skips FF5.4, skips characterizations, skips formatting; skips everything. This model is so ass bro.
Gemma-4-31b (paid) — I like this model, I’m only putting this here because the anti-slop measures MURDER this model.
Mimo 2.5 Pro — hard refusals. Notoriously end point sensitive; requires specific settings.
CURRENTLY WORKING ON:
Trying to iron out the random occurrences of named characters gaining tails or other non-canon traits, without sacrificing randomized NPC depth.
Writing more starting scenarios. Hopefully.
Authors notes:
V.1: 08/30/2026 — release
V.2.0: 09/03/2026 — HUGE speed ups and optimizations (15k permanent tokens down to 8k). Added additional lorebooks for more content.
V.3.0: 09/09/2026 — Dead Dove improvements; character traits adhere much better (GLM5.2), better prose (I think), much better sense of conversation dynamics. Interactions are less ‘robotic’ and ‘game-y’ — far more engaging in my opinion.
Utworzono 29 cze
Zaktualizowano 26 wrz
Opublikowano 30 sie