Local inference is now available!
Oleg Skuttesteam_community_announcements
It has been just over two months since the game's release, and I have closely monitored player feedback. The vast majority of the criticism has been related to two main issues:
- In-game currency
- Censorship
To address these concerns, I have added support for local inference. From now on, you will be able to run the game using only your own hardware. This allows you to use any LLM, Stable Diffusion, and TTS models, and also eliminates the need to pay for cloud services. Moreover, it's fully self-contained, so you don't need to install Python, separate server executables, or WebUIs (although support for external API endpoints is coming soon too).
Text Generation
I have integrated llama.cpp directly into the game. This way, I get closer to the actual models and have full control over sampling. It runs on Vulkan, so no NVIDIA GPU is needed. It accepts models in .gguf format.Image Generation
Also integrated directly into the game with no dependencies. Runs on CUDA, so NVIDIA GPU is preferred. It accepts models in .ckpt, .safetensors, .diffusers, and .gguf formats.Speech Synthesis
I have integrated two options:- VITS: Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech
- JETS: Jointly Training FastSpeech2 and HiFi-GAN for End to End Text to Speech