Fluff Preset

PRESET RELATED INFORMATION

Main V1:

• Made to watch Gemini try to output more enjoyable writing.
• Two thinking models: one will try do to its best at first attempt, other will draft and improve before outputting the message. The short one should take 500-750 thinking tokens, but Gemini does what Gemini wills so sometimes it triples that.¹ Speaking of, the latter will consistently take 2000+ tokens. Only you can decide whether longer waiting times are worth it.
• The U3 prompt contains rules, the U5 prompt contains thinking instructions. Edit them as you see fit.

¹I guess you could play with Max Response Length / Reasoning Effort if you both really want to keep it short and never plan to use the longer one.

It was superceded by Main V2, but the download link, if you want it for whatever reason, is here.

Main V2:

• I think, on average, it writes better. Still has its moments, still heavily dependent on context, still bad at NSFW¹, but across couple chats I'm happy with what I saw.
• I gave up on trying to thinking instructions besides a short reminder at the start of the context and the prefill. It's now a bit of a roulette on how much it will think. It isn't too bad. Also, the 'forgets to close the <think> tag' thing is mostly fixed now.
• Added 3 PHI prompts, one for 'the context is dogshit please write better' (limited success, but does something I believe), second for 'please stop with the melodrama and just give me cute girl who loves me action', third for any other instructions you might want it to follow (add more dialogue, force it to push the story forward, etc etc).

¹I'd say that Gemini isn't godawful at it, especially if you have coom-poisoned chat, but only if you write a whole separate preset focused on sexo. Trying that with this one wouldn't work, I think.

Experimental:

• Made to watch Gemini go nearly schizophrenic while thinking. See this gallery to get a general idea.
• On (near) empty contexts it will fuck up its thinking box, for whatever reason.
• The U3 prompt contains rules, the U6 prompt contains PHI. Edit them as you see fit.

HOW GOOD IS IT?

In general, Gemini really likes context poisoning. Fancy greetings and defs = fancy writing.

Main: Nice, but don't expect miracles if you put it in a 20k context chat full of -isms. It will try its best, but its best won't be enough.

Experimental: worse writing in my opinion, more prone to -isms, but can be a better pick if you prefer simpler prose.

Neither of them are probably any good at NSFW (I have not tried that), partly because Gemini in general is bad for it, partly because there are no instructions for it included. You could add yours in the PHI for example, but I advise just using Deepseek for lewds. Less headache that way.

CHANGELOG

2025/08/22 - updated Main to V2.
2025/07/18 - added a new version.
2025/06/29 - changed everything into something more experimental.
2025/06/16 - another update.
2025/06/07 - small update.
2025/06/06 - changed pretty much the entire thing. Old Link: here.
2025/01/14 - even more changes, I kinda forgor what exactly besides changing the main prompt. Should be the final update until 2 Pro drops.
2025/01/09 - debloat, added support for example dialogue, changed the ero prefill a bit. Not 100% sure how it will perform so here's the earlier version.
2024/12/17 - rewrite. Should work better now.
2024/12/09 - updated for 1206.
2024/11/30 - changed some little things, added an optional ero prompt. I'm an 'ah ah mistress' kind of guy so don't expect miracles when it comes to NSFW quality though. Remember you can edit it with your own stuff.
2024/11/29 - small update. Should fix an error where it repeats <previous_message>. Also some small adjustments that should reduce {{char}} repeating what you said word for word.
2024/11/29 - changed the preset from Claude to Gemini. You can still download the former here.

click for cute and fluffy bots