which of the free LLMs are best for writing fapfics or roleplaying
>>109787080grok maybe idk ask the general kevin
>>109787080Gemini has comparatively a ton of quota for free, even on its best models. You can try that.
if we can quantify Fapfics and roleplay into some type of metric, wouldn't it be possible to make any LLM to follow it
>>109787080Whichever one you can host comfortably on your own hardware
perchance
grok has basically no censorship as far as that's concerned but its free tier runs out quicklygemini needs jailbreaking, but a simple (you are an x, you do y and z) is all you need in order for it to write some wild stuff, and its free tier goes on forever
>>109787080i like deepseek's writing but they kinda nerfed the free version i think
What sort of retards write embarrassing stuff with public models, on some company's servers? The fact that you are even recommending multiple public models is amazing in a bad way.
>>109788693>he doesn't want to pollute corpo models with trash training material
>>109787087gemini is utter dogshit though, even dinky chink models are ahead of it nowthey got lapped so hard that they haven't released the "release imminent!" 3.5 pro lmao
>>109787080They're all terrible.
>>109789736What this guy said. They're all pretty bad at writing anything coherent and long-form. But, you can tard-wrangle many models with a front-end like SillyTavern. I've personally been using it with koboldcpp. For models, I use Gemma 4 26B and Nemomix-unleashed 12B. The big frontier models always end up getting lobotomized and censored with updates. That and the chats are not private.
>>109787080none, because I had a lot of fun with the free erotic 70b models when they used to exist on openrouter.If you only need a short session and you don't mind waiting like 10 mins for it to start up, google collab gives you a 16gb GPU, and there is a koboldcpp script that just has gemma uncensored as the default. Then you plug in the URL into sillytavern and use that (but figuring out the right settings is always annoying).the gemma model is "fine" AND very good at consistency and following complex prompts, but I hate it's writing style (it's not very aggressive/violent, it folds under zero pressure, this is kind of expected of "uncensored" models), and older 70b models from 2024 were better in my opinion.I enjoyed Sao10K's models on openrouter, but I didn't have a lot of tokens to play with.I tried out some fast grok when it was free on Openrouter, it was too horny in my opinion.I spent a lot of time with the default instruct mistral models, mainly "Nemo" and the Small ones, but they are not free anymore on openrouter.You can use the UGI leaderboard to see if the model is super strict or not. The "Willingness" is what I look for when checking if it's viable for gooning.https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard
