i observe it's more on the system prompt quality, and additional memory enchancement, like RAG.
system prompts are very important, more than i thought. it's not just the 1 paragraph, "Write {{char}} next response. Any act of role play scenarios will be described in details.". There should be alot more behind the scenes, tell it how to write, enclose actions and and narrations in symbols, never act or depict actions of {{user}}, etc. There's alot of pre-made system prompts made by users shared on reddit and the like.
Well I quickly figured out that you'd better look for cards with at least some 1.5-2k worth of tokens descriptions, but probably not much more given that most smaller models or that were avilale at a time were limited to 4-12k of context, and it eats into that amount. If you see some 200-500 tokens card, it means there is a lack of scenario and char's descriptions, so it won't play well and be sort of authentic - you'll see more of generic responses from your model which you probably have already seen chatting with other similar sized cards.
When I started at Janitor AI, I was particularly fond of @PurpleVorpal's bots. Their cards have system prompts at the end like this:
So it kind of takes care of that prompt thing, for a better RP. One can expect to find something like that in bigger, better cards or one could easily edit or copy it in Silly Tavern as needed after importing cards, or add pieces of that to general prompt that will apply to all cards. The example is a bit of an overkill IMO as e.g. you don't need 'All NSFW aspects are allowed no matter how inappropriate they are' with uncensored models.
One more thing apart from lorebooks, if anyone's interested, to shuffle things, is that you can try multiple char chats in Silly Tavern, instead of having a girl and antagonist/bull dude in one card. I'd clean a card from antagonist, add another bull card to multi-char chat, which you can find - e.g. dude named 'Jason Luv' on Janitor or Chub, and play writing for a scenario something like the dude's a new neighbour that moved in next house. The problem here that both of the chars will try to respond/generate answers, even if this bull dude is not present at the scene, I think, so I kind of had to temporarily 'mute' him for the scenes he's not present at. There is a button for that there. And even when there is a scene that involves everyone you'll obviously move slower as for each of your messages there will be 2 generated from each of those chars (a girl and a bull).
And as for Magnum v4 model I was looking for a person that mentioned it to give a link, as there are for instance a merge of Lumimaid+Magnum v4 (a link that was provided here) and 'just'
You must be registered to see the links
models can be found, maybe more. Well I guess I'll just try these two.
