[Build Help] Best RP models and frontends for 4090 (24GB VRAM) / 64GB RAM? (No SillyTavern)

Posted by Ok_Cartographer_809@reddit | LocalLLaMA | View on Reddit | 8 comments

Hi everyone,

I'm looking for some recommendations to level up my local RP experience. My current setup is a Windows machine with an i7-14700K, 64GB DDR5 RAM, and an RTX 4090 (24GB VRAM).

I am currently using LM Studio, which I like for its ease of use. However, I’m looking for a frontend that is more specialized for Roleplay—specifically something with robust support for Character Cards and Memory/Lorebook features—without going down the SillyTavern rabbit hole.

For models, since I have 24GB of VRAM and plenty of system RAM, what are the current "S-Tier" recommendations for high-quality, creative RP in 2026? I’m interested in models that:

Excel at nuanced prose and avoiding "GPT-isms."
Can handle long-context roleplay without losing character consistency.
Fit well within my hardware (I'm open to GGUF or EXL2).

Questions:

Is there a frontend that bridges the gap between LM Studio's simplicity and SillyTavern's features? (e.g., Faraday/AnythingLLM/etc.)
Which 30B-70B models are currently the favorites for immersive storytelling on a single 4090?

Thanks for the help

[-]

Silent-Spaz259@reddit

Those suggesting to "vibe code" a system are ignorant to the requirements of what you're asking. A slop machine won't be able to hallucinate a competent solution.

Try search GitHub for "character AI" or "AI roleplay". There are some ok alternatives.

Lorian0x7@reddit

Skill issue, or you must be the ignorant here. I vibe coded my custom frontend with no issues and it feels a lot better than Silly Tavern.

Sure you have 👍

🖕

🤣 thanks for the laugh pal.

FinBenton@reddit

I'w replaced everything with gemma-4 31b on 5090, 26b moe is very good too and easier to fit to 24GB.

Personally I have vibe coded a few llama.cpp wrappers for backend, I can definitely recommend spending a couple of days to just code a web ui around llama.cpp yourself, its easy and you get something that works much better because you can personalize everything.

This. I did the same thing. Somehow models inside Silly Tavern always felt broken. Building your own frontend is the best advice here

Background-Ad-5398@reddit

why dont you just drop that prompt into your favorite sota model and have it build you a frontend