I have been making videos, and having a lot of local AI shenanigans running on my 3080 that most claim need a 40 at least.
I am surprised no one made a Ollama type model for AI dungeon’s Voyager thing. Voice modulations don’t need even a GPU from 2010s, Image generation is quite good even on low B models. The only thing that is slow is the model generating the story but most Websites use the cheep ones I am using Qwen’s Q8 model on a I5 CPU and it takes about 10-20 min with max context. Q4-5 take like 2-3 min
Even if it’s not a Open soruce base you can make a high quality text based RPG out of that.
Edit: KoboldCpp I just needed to use slop GPT to find it google is just bad. Still a good post to keep
Have you tried asking in !localllama@sh.itjust.works?

