ReadyArt's Wicked-Tutu-31B Brings Private Mature Roleplay to Local Hardware
ReadyArt drops a 31B uncensored Gemma-based roleplay model in GGUF format, built for explicit fiction, horror, and violence on local hardware.
- ReadyArt released Wicked-Tutu-31B-v0.1-GGUF, a 31B uncensored roleplay model in GGUF format.
- Built on a Gemma-family base, tagged for ERP, horror, and violent fiction.
- Apache 2.0 license, runs via llama.cpp, Ollama, LM Studio, vLLM, and Docker Model Runner.
- Already past 6,000 downloads despite minimal social engagement around the release.
- No model card, dataset, or benchmark details are published alongside the weights.
- Part of a rapid ReadyArt release cadence spanning v0.1, v0.2, v0.25, and v2.0 variants.
Wicked-Tutu-31B packages mature roleplay for local inference
ReadyArt has released Wicked-Tutu-31B, a 31-billion-parameter text-generation model for mature creative writing and roleplay. Its GGUF packaging supports local inference through llama.cpp, Ollama, LM Studio, and compatible tools, allowing prompts and outputs to remain on the user’s machine.
The repository carries tags for explicit content, erotic roleplay, horror, violence, conversation, and reduced-refusal behavior. It accumulated several thousand downloads within its first days despite limited public discussion. ReadyArt is a prolific publisher in this category, with nearly 400 releases listed on its publisher page at publication time.
A 31B Gemma-family build
Repository metadata labels the architecture gemma4, indicating a Gemma-family foundation. ReadyArt does not identify the exact base checkpoint or publish enough training information to verify the model’s lineage, so runtime compatibility should be checked against the metadata rather than inferred from the name.
| Field | Details |
|---|---|
| Parameters | 31 billion |
| Distribution format | GGUF quantizations |
| Suggested quantization | Q4_K_M |
| Primary use | Mature roleplay and narrative generation |
| License shown | Apache 2.0 |
GGUF stores quantized weights that trade some numerical precision for lower memory use. A 31B model at Q4_K_M generally requires about 19GB for the weights, while the context cache and runtime buffers increase total demand. A 24GB GPU can handle modest contexts, and
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
