Apparently, private Local AI roleplaying AI is a pretty serious use case. ReadyArt's new Wicked-Tutu family is already pulling serious downloads. And this is not a tiny model Stats 🧠 31B dense Gemma 4 💾 Q4_K_M → ~18.7GB 🎮 fits in 24GB-GPU territory 📚 up to 256K context 🏠 completely local Runs directly with ⚙️ llama.cpp 🦙 Ollama 🖥️ LM Studio ReadyArt trained it specifically as a private local roleplay model, with a purpose-built dataset and thousands of scenario categories. The earlier v0.1 GGUF has already passed 17K downloads, and they're iterating fast enough that a Final 31B build is already out. Local AI isn't only competing with ChatGPT but it's enabling categories where privacy + customization + no cloud at all are the product. ⚠️ The HF metadata says Apache 2.0, but the model card also adds personal/non-commercial terms, so I'd treat the licensing as unclear for now. Link in ALT