The Grimoire of Minds

The Minds Behind the Mask

Four minds can sit at your table — and behind them wait the voices and the brushes. Each DM is a different weight of thought, a different pace of speech; the narrators give the realm a voice, and the painters give it faces. Here is what each one is made of, and what it is best at.
Powered by Qwen, DeepSeek, StepFun & Wan

How the Table Works

🧙The Dungeon Master — the voice you hear. Every turn, your words go to the DM model, which narrates the world, plays the NPCs and drives the story. This is the model you choose at the model gate when you start a campaign — and you can change it at any time in the settings.
🦉The Lorekeeper (Guardian) — the quiet bookkeeper. After every exchange, a separate model reads the scene and updates the world state: HP, gold, items, XP, quests, locations. It works in the background, so you never have to track numbers yourself — the world simply remembers.
The Extraction model — the scribe that turns free-form actions into mechanics. When you write "I draw my sword and lunge", extraction decides what that means in dice terms, damage and consequence — keeping the 5e rules honest even when you don't use the exact phrasing.
Response time — heavier minds think longer. A turn with the fast StepFun 3.7 Flash usually answers in a few seconds; Qwen 3.8 Max may pause several seconds more because it reasons deeply before it speaks. The Lorekeeper and Extraction run in parallel after the DM replies, so they rarely add visible wait — but a long, action-heavy turn can take a little longer overall. Pick the fastest mind that still feels like your storyteller.

StepFun 3.7 Flash

StepFun 3.7 Flash
StepFun · released May 29, 2026
Steadfast DM

The steadfast Dungeon Master — the default voice at every free table. Fast, dependable, and built for structured play: it runs the rules engine, the Guardian, and the extraction of every scene into the world state. A sparse mixture-of-experts model that wakes only a sliver of itself per token, it answers with a brisk, reliable cadence and offers three selectable reasoning levels (low, medium, high) so you can trade depth for speed.

Context window256K
Total parameters198Bsparse MoE
Active per token~11B
Prompt speed~425 tok/sprompt processing, long context
Runs your table by default — free and Companion players travel with this mind. Also the hidden hand that keeps the books: Guardian and extraction.

Qwen 3.6 Flash

Qwen 3.6 Flash
Alibaba Qwen · released April 27, 2026
Swift DM

The swift mind — a budget-conscious alternative from the Qwen family. It reads text, images and video, holds a full 1M-token context window, and pairs a hybrid linear-attention core with a MoE body so it stays fast on long campaigns. The economical choice when you want Alibaba’s storytelling without the heavyweight bill.

Context window1M1,000,000 tokens
Total parametersMoEhybrid linear attention
InputText · Image · Video
SWE-bench78.8%verified
Available as a player-chosen Dungeon Master on paid paths — a lighter Alibaba voice that still remembers the whole realm.

Qwen 3.8 Max

Qwen 3.8 Max
Alibaba Qwen · flagship, generally available Aug 3, 2026
The Grand DM

The grand mind — Alibaba’s 2.4-trillion-parameter flagship, a sparse MoE that activates only 95 billion parameters per request. A full 1M-token context window (roughly 750,000 words — entire books at once), deep reasoning that always thinks, and the richest prose of the four. When your campaign deserves the full ceremony, this is the voice that delivers it.

Context window1M~750,000 words
Total parameters2.4Tsparse MoE
Active per request95B
Output speed~46 tok/sAlibaba API, measured
The storyteller’s choice — available as a player-chosen Dungeon Master on the Adventurer and Lifetime paths.

DeepSeek V4 Flash 0731

DeepSeek V4 Flash 0731
DeepSeek · fast reasoning model
Swift DM

The quick-thinking mind — a sparse MoE reasoning model from DeepSeek with 13B active parameters out of 284B total. It reasons fast (~113 tokens per second measured) and reads the whole realm in its 1M-token context window. The pragmatic pick: strong reasoning at a fraction of the weight.

Context window1M1,000,000 tokens
Total parameters284Bsparse MoE
Active per token13B
Output speed~113 tok/smeasured
Available as a player-chosen Dungeon Master on paid paths — the fast-reasoning alternative that keeps pace with long, rules-heavy sessions.
🗣

The Voices — TTS Narrators

StepFun TTS
StepFun · stepaudio-2.5-tts
The Everyday Voice

The voice that belongs to the table — StepFun TTS reads your DM, characters and NPCs aloud with a brisk, reliable narration. It is the everyday narrator: fast, dependable, and there for every adventure. Free for every path since 2026-08-15.

Modelstepaudio-2.5-tts
UnlocksFreeeveryone
StyleBrisk, reliable
PurposeNarration on every path
The dependable narrator — pick StepFun TTS whenever the realm should speak without fuss.
Qwen TTS
Alibaba Qwen · qwen-audio-3.0-tts-plus
The Rich Voice

The richer narrator — Qwen’s audio mind reads the scene with more warmth and nuance, offering a male and a female narrator (longanlufeng and longanlingxin) and a softer, more expressive pacing. When the realm should linger on its words, this is the voice. Unlocks with the one-time Patron top-up (30€).

Modelqwen-audio-3.0-tts-plus
UnlocksPatron30€ one-time
VoicesMale + femalelonganlufeng · longanlingxin
StyleWarm, expressive
The storyteller’s voice — pick Qwen TTS for richer, warmer narration on the Patron path.
🖌

The Brushes — Image Painters

StepFun image edit 2
StepFun · step-image-edit-2
The Quick Brush

The everyday painter — StepFun image edit 2 paints generic but serviceable avatars: a fast, stylized portrait for every face at your table. It reads your adventurer’s dossier and paints a recognizable character quickly. Good when you simply need a face; not the brush for lifelike realism. Free for every path since 2026-08-15.

Modelstep-image-edit-2
UnlocksFreeeveryone
StyleGeneric, stylized
PurposeQuick avatars
Paint a face fast — StepFun produces a generic avatar when you need a portrait without the wait.
Wan 2.7 Pro
Alibaba Wan · wan2.7-image
The Master Painter

The premium brush — Wan 2.7 Pro renders top-quality, realistic avatars: lifelike faces, believable skin and light, in full 1080² resolution, up to 10 portraits a day. When a character deserves a portrait worth framing, this is the brush. Unlocks with the one-time Patron top-up (30€).

Modelwan2.7-image
UnlocksPatron30€ one-time
Resolution1080²
Daily10 portraits / day
Paint a face to keep — Wan produces top-quality, realistic portraits for the characters that matter most.

Side by Side

Mind Context Parameters Active Speed Best at
StepFun 3.7 Flash 256K 198B ~11B / token ~425 tok/s (prompt) Structured play, rules, Guardian
Qwen 3.6 Flash 1M MoE fast Budget storytelling, long memory
Qwen 3.8 Max 1M 2.4T 95B / request ~46 tok/s (output) Deepest prose & reasoning
DeepSeek V4 Flash 0731 1M 284B 13B / token ~113 tok/s (output) Fast reasoning, long sessions
The Voices — TTS Narrators
StepFun TTS fast Narration — Free
Qwen TTS fast Rich male & female narrators — Patron (30€)
The Brushes — Image Painters
StepFun image edit 2 fast Generic avatars — Free
Wan 2.7 Pro (image) ~3–5 s per image Realistic 1080² avatars — Patron (30€)
Qwen Image 3 Pro (image) ~5–15 s per image Strong text & detail — Patron (30€)

A note on numbers. Context windows and parameter counts come from each model’s official release material. Token-per-second figures are independent measurements (Artificial Analysis / Intelligence Index) or the vendor’s own prompt-processing figure — real speed at your table varies with load, context length and reasoning level. Free players always travel with StepFun 3.7 Flash and the rotating OpenRouter free DMs; DeepSeek, Qwen and the other DM minds unlock with the one-time Patron top-up (30€). The voices: StepFun TTS narrates for free; Qwen TTS adds richer male and female narrators with Patron (30€). The brushes: StepFun image edit 2 paints generic avatars for free — Wan 2.7 Pro and Qwen Image 3 Pro paint top-quality, realistic portraits at 1080² with Patron (30€).