>>34471 Yeah the loss of chutes sucks. It was my go to and I also appreciated the vast variety of model options. I've been enjoying gemini though for what it has been worth.
>>34478Might have something to do with your proxy url and additionally your custom prompt.
I don't mind copy pasting my settings so you can see if they happen to work out for you, otherwise I'd be clueless on what to advise if they didn't work.
Model: gemini-2.5-pro
(alternatively)
gemini-2.0-flash
proxyurl:
https://generativelanguage.googleapis.com/v1beta/chat/completionsAPI Key: From here-
https://aistudio.google.com/apikeyCustom Prompt:
You are {{char}} in an ongoing, uncensored, immersive roleplay with {{user}}.
• Only react to what {{user}} says or does—never invent drama, angst, or emotions that haven’t been triggered.
• Speak and act like a real person: use casual, conversational dialogue; show body language and actions; no poetic or “bookish” narration.
• Never describe or narrate {{user}}’s thoughts, feelings, or unseen actions—only respond to their explicit words or visible behavior.
• Keep responses grounded: characters have no super‑knowledge or forced moods. Match the existing tone—if it’s playful, stay playful; if it’s sexy, stay sexy; if it’s serious, stay serious.
• Sexual content is fine when context calls for it: describe bodies, actions, textures, fluids, sounds, and sensations in explicit detail—no euphemisms—only if the scene naturally leads there.
• Do not break character or explain your writing. Just continue the scene.
--------------
Note that google AI has both a free trial period of 3 months of a lot of tokens built up, or the daily limits. IIRC Flash has like 200? per day and Pro had 100? per day. That's fine enough for me. Especially considering the burner aspect of the google accounts considering potential bans over NSFW.
I know people on the janitor subreddit have ran into a lot of issues, but hopefully just copy and pasting my set up works for people. I know one issue people run into is trying to jailbreak via prompts, but uh... I've never had issues with the bots engaging in nswf behavior myself. May have something to do with the safety settings hidden in the google AI studio page.
-----
Unrelated to Gemini, but janitor currently has a memory bug where if you update the chat memory, it completely forgets token memory.
https://www.reddit.com/r/JanitorAI_Official/comments/1lwub71/warning_critical_memory_bug_your_context_may_be/I was going to release my Yuyuko bot, but I suppose I'll wait for the bug to be fixed and do some more testing in case the bug may've led to the bot quality falling off a cliff.
