CoolFace
Modelpublic

nohurry/sillytavern

sourceHugging Facemitupdated 2d agoView on Hugging Face
22likes
Model Card

SillyTavern

Publicly shared presets, lorebooks, characters and whatnot.

Presets

Moonlight (v1.0)

Preset download: link

A simple chat preset which helps improve Gemma 4 31B's prose and adds randomization to swipes.

Tested with Bartowski's Gemma4-31B-IT Q6KL, 32K Q80 context. Model download: [link](https://huggingface.co/bartowski/googlegemma-4-31B-it-GGUF/blob/main/googlegemma-4-31B-it-Q6K_L.gguf)

It might work with Gemma4-26B-A4B and other non-Gemma4 models.

Voyage (v1.0)

Preset download: link

Based on Moonlight, I turned it into a Powered by the Apocalypse (PbtA) "lite" experience. The idea here is PbtA systems have high-quality prose and a really simple yet effective decision-making framework. By just using it's core, Gemma4 is already writing much better interactions and dialogue.

  • —Added GM notes functionality
  • —Added tracking notes
  • —Added a tag system (akin to Dungeon World's tags and Pathfinder 2e's traits)
  • —Improved NPC creation (now using tag system)
  • —Improved location creation (now using tag system)
  • —Improved writing instructions edge-cases
  • —Improved model affirmations
  • —Streamlined naming rules
  • —Removed Chain-of-Thought (Gemma does it much better by default)

Tested with Bartowski's Gemma4-31B-IT Q6KL, 32K Q80 context. Model download: [link](https://huggingface.co/bartowski/googlegemma-4-31B-it-GGUF/blob/main/googlegemma-4-31B-it-Q6K_L.gguf)

It might work with Gemma4-12B, Gemma4-26B-A4B and other non-Gemma4 models.

Voyage (v2.0)

Preset download: link

Based on Voyage, I tried to improve the roleplaying experience by introducing cause-and-effect.

  • —Bind GM to assistant role instead of "you" to prevent talking as User
  • —Consolidated NPC / Location / Scenario generation
  • —Improved tag-based generation (uses cause-and-effect)
  • —Improved PbtA Core skill check triggering
  • —Use scenarios instead of events, with clear definition
  • —Improved writing instructions to reduce slop (staccato overuse, "not x but y")

Tested with Unsloth's Gemma4-31B-IT QAT Q4KXL, MTP enabled, 32K BF16 context. Model download: link

It might work with Gemma4-12B-IT QAT, Gemma4-26B-A4B-IT QAT and other non-Gemma4 models.

Voyage (v3.0)

Preset download: link

Rebuild from the ground up.

  • —Reduced token usage
  • —Reworked skill check
  • —Improved backstories
  • —Improved NPC dialogues and interactions between NPCs
  • —Improved prose
  • —Fixed many issues.

Tested with MTP enabled, 32K BF16 context on the following models:

  • —Unsloth's Gemma4 31B IT QAT: link
  • —Unsloth's Gemma4 26B-A4B IT QAT: link
  • —Unsloth's Gemma4 12B IT QAT: link
  • —Unsloth's Gemma4 E4B IT QAT: link
  • —Unsloth's Gemma4 E2B IT QAT: link

It might work with other non-Gemma4 models.

Eval (v1.0)

Preset download: link

A playground of various rules to experiment with. Meant as a development resource over normal use.

Tested with Unsloth's Gemma4-31B-IT QAT Q4KXL, MTP enabled, 32K BF16 context. Model download: link

Might work with other non-Gemma4 models.

Voyage (v4.0-exp)

Preset download:

Experimental releases.

Tested with Unsloth's Gemma4-31B-IT QAT Q4KXL, MTP enabled, 32K BF16 context. Model download: link

Might work with other non-Gemma4 models.