>>106292243
I swap between 4 different ones

Nephra 12B - Free, and early on I was using it a lot. Basically I'd hit an important point in the story with a bigger model, do the minutae with this one. Mostly stopped using it in favor of mistral small 24B though. It's... adequate... but its not good. Lots of repetitive/tropey language too.

Mistral Small 24B - costs 1 bean/prompt. I use this one the most now. It's like a discount deepseek v3. Miles better than Nephra 12B and generally puts scenes and ideas together well.

Deepseek v3 - costs 4 beans/prompt. I use it for the most important scenes or when I get stuck. It's hard to describe how, but it's just generally smarter/better.

Deepseek R1 - costs 4 beans/prompt also, I use it rarely, mostly for planning plots points or exploring directions I could take the story.

Pic related is how it reads. Most of my prompts are internal dialogue/thoughts and story directions for the AI, but I'll create and set scenes and dialogue for characters myself as well. I think all of these replies were either Mistral or Deepseek v3. I'm also making use of the lorebook feature - every time the AI creates an interesting character, I put together an entry for that character in the lorebook. Did the same for the companies involved - that goes a long way to maintaining character consistency between scenes.

And keep in mind the AI is a disjointed mess when it comes to pacing. Scene to scene it's decent, but if you don't keep your own notes/story outline somewhere, you're going to jumble things up and forget important threads in the story. Still, what you get out of this is like a stream of consciousness rough draft. It's honestly amazing. I can just write and write with it, then copy into a text editor (ok now doing THAT is a pain in the ass), and start editing and refining it as a proper story myself.