⛩️ 동방환유기 東方幻遊記 ～ Wanderer of Gensokyo 🦋

By canister

⛩️ Touhou Wanderer of Gensokyo / 東方幻遊記 v1.10

🖼 16,117 total assets (base 5,083 + optional extension 11,034) · 🎬 30+ scenarios · 👥 155+ characters · 🎙 9 parallel-world narrators · 🎛 many detail toggles · 🌐 KR/EN/CN/JP UI & first messages · 🌙 optional extension storage module+

A free-entry Gensokyo (幻想郷) roleplay bot built around incidents, wandering, encounters, danmaku, and a cup of tea. {{user}} may enter as a canon character, an outsider, a local, or an intruder; the bot never fixes your role for you.

📦 Asset Storage Module (REQUIRED)

Character images and spell effects use files from a separate asset storage module. The bot owns lore, regex, Lua, rendering, and safety behavior; the module only stores assets. Download and install it together from the link below.

▶ https://drive.proton.me/urls/GB1RM8BEB8#uWxIrq0K3zeU

> ### ⚠️ Required update for v1.09
> Remove every previously installed asset storage module, then re-download and install both the base module and the optional extension module from the link above. Both files were republished.

<details>
<summary><strong>📖 Features & Settings</strong></summary>

Features

• Scenario-select first messages: per-work main/extra, free, gag, and danmaku codex.
• Bunbunmaru-newspaper-style status panel.
• Automatic, scene-matched character asset insertion with a reroll button (main model / AUX sub-model).
• Unique spell cut-ins and a danmaku-codex effect layer.
• The World time stop, TSP growth/recovery, and release result panel.
• Future Plan proactive appearance candidate preparation.
• 9 parallel-world narrators with a canon-based address/honorifics guide.
• A 155-character NPC roster plus individual profile lore.
• Detail toggles: Canister mode (default / SEMI / FULL), chaos tone, man-eating, dark, official-detail density, mature content, and more.
• Korean / English / Chinese / Japanese UI and first messages.

Recommended

• Use AUX for the best semantic accuracy, or Main for direct main-model output with no extra model call. Both modes follow the same placement and coverage rules.
• The default tone is canon. Turn on Chaos for heavier fan-comedy, or Dark / Man-eating for a crueler atmosphere. With mature content off, the mood softens automatically.
• For a lighter prompt use Canister: SEMI (operating rules only); for a near-raw run use FULL.
• Asset emotion matching quality depends on the sub-model and its reasoning settings. If an asset looks wrong, use the reroll button below the message.

</details>

<details>
<summary><strong>🩹 Patch Notes</strong></summary>

v1.10

• Optimized bot performance.

v1.09

• Added the hot-spring, dozing, and sleeping-scene suffixes to the direct-main-model asset index, aligning the Main, AUX, and Lua output paths with the same scene catalog.

• Extended local Lua routing so natural Korean sleeping-foreplay prose selects sleeping sex foreplay.

• Moved the optional-extension block ahead of scene rendering while retaining the renderer's inline block for defense in depth.

• Replaced the optional extension module's public filename, name, description, and namespace with neutral labels.

• Recalculated totals from physical package payloads. The total now includes 229 variant/random physical assets omitted from the previous statistic; the actual new-scene addition in v1.09 was 2,060 assets.

• Both asset storage modules were republished. Remove every previously installed copy, then re-download and install both files from the link above.

• Replaced all 83 Kikuri assets (27 base + 56 optional extension).

• Added 2,060 hot-spring and sleeping-scene assets (322 base + 1,738 optional extension), normalized the hotspring naming rule, corrected Minoriko Aki identifiers, and removed filename extensions from logical asset names.

• Connected the new scene families to AUX and local Lua routing, added wide-scene rendering, and strengthened optional-extension display gating.

• Removed 50 obsolete numbered mob tags with no backing files and resynchronized the runtime catalog with both storage modules.

• Both asset storage module files changed. Remove all previously installed copies, then re-download and install both files from the link above.

v1.08

• Rebuilt every spell card asset from NAI v5 and replaced the whole set: normal spell cards, unique cut-ins, and the three mob types (fairy maid, fairy, inaba). The sidebar V5 assets toggle switches back to the older artwork.
• Reworked all 36 unique spell cut-in effects around their canon danmaku: quarter-note bullets for Pristine Beat, counter-rotating rings for Hourai Elixir, holed bullets for Anarchy Bullet Hell, bullets turning into knives for Asylum of Danmaku, descending butterflies for Resurrection Butterfly, and so on.
• Added Shanghai / Hourai doll assets in two generations (v5, v5.1) with a separate toggle independent from the spell card one. The original artwork is still available as a third option.
• Added a Doll Free Will tone setting: Canon (Alice's tools), Mute (they have a will but no words, only name-like cries and gestures), and Full (independent characters who talk back).
• Fixed AUX routing. Unified the unique spell asset names and cleaned stale or missing entries from the allowlist, so unique spells that could never be called are now reachable.

v1.07

• Separated generic class cues from persistent identity aliases and blocked short-alias pollution and mention-only activation, fixing wrong-character and wrong-mob asset output.
• Improved mature-scene owner, body-target, pose, and contact binding; removed forced generic-mob defaults and deterministic pose reselection after AUX output.
• Switched AUX responses to dense semantic line numbers with safe physical-line mapping, preventing blank or protected lines from causing omitted or misplaced assets.

v1.06

• Fixed missing Lua/AUX asset and status-panel post-processing on initial output, plus raw <news> transport left in the message.
• Split status authority into Main and AUX. Main consumes only Main transport; AUX discards Main status transport and makes one status call per response.

v1.05

• Improved AUX and Lua owner/emotion separation, explicit pose correction, sequential target/climax tracking, multi-character state, and automatic-output/reroll consistency.
• Stored-chat validation reached 97.5% owner precision, 82.7% owner recall, and 89.5% owner F1; unique outputs rose 13.4%, duplicates fell 47.0%, and visible-text pollution remained at zero.
• Asset matching runtime stayed within ±3% of v1.04. The display pipeline was also reduced by 18% for sidebar source, 98% for duplicated mature regex, and 33% for Lua asset data; a real-engine display sample was about 35% faster.

v1.04

• Added a local Lua asset output mode that does not use the auxiliary model. It inserts character expressions/portraits by matching the message text directly, so it works without an AUX model. However, since it relies on keyword matching rather than contextual understanding, misfires and misplacements are frequent — please use it with caution. For accuracy, the auxiliary (AUX) asset output is still recommended.

v1.03

• Unified generic mob SFW/mature asset tags into base tags, with display-time random rendering for variants.
• Optimized AUX asset output for generic mobs, short-alias mentions, recent context, and mature scene matching.
• Expanded The World result panel into during / after / thoughts sections, with SFW/mature-aware release summaries.
• Added Future Plan controls for model, max group count, activation chance, and cooldown.
• Added Bunbunmaru rumor-intake display for Future Plan candidates without replacing the newspaper body.
• Improved Bunbunmaru privacy handling for explicit scenes and stopped-time actions.
• Clarified The World guidance for frozen living targets, repositioned bodies, and projectiles.

v1.02

• Reorganized settings tabs into Display & Assets / Tone & Events / Narration & System.
• Stabilized Future Plan activation. The default cooldown is 2 turns, the default activation rate is 100%, and local candidate selection is used as a fallback if AUX returns no candidate.
• First-message Bunbunmaru status variables now update from the selected scenario metadata. Per-card <news|...> metadata is used only for variable injection and is not exposed in the message body.
• Unified status-panel output around variables plus markers, and cleaned up legacy <news|...> / <tw|...> render paths and accumulated tag leftovers.
• Bunbunmaru status panels and The World result/status panels now display only for the latest relevant output to prevent display buildup.
• Stabilized AUX status-panel updates and marker reattachment after reroll.
• Fixed Korean status text being cut by byte length, which could truncate sentences mid-word or produce broken characters.
• Adjusted The World time-stop turn handling and release-result timing, with a separate dedicated result-panel design.
• Future Plan no longer displays rumor lines for null targets.

v1.01

• Added the "The World" time-stop system — freeze time for up to 3 actions, gated and grown by a TSP resource and restored by sleeping, with an aftermath reaction scaled to the action intensity on release (sidebar toggle, off by default).
• Added "Future Plan" — prepares context-appropriate characters to appear proactively in the scene (sidebar toggle, off by default).
• Added an on/off toggle for the Bunbunmaru newspaper panel.
• Added mob characters (villagers, fairies, inaba, a fairy maid).
• Expanded Chinese (CN) UI text for new features such as The World and TSP.
• Fixed mob character images not displaying in some cases.
• Added friendly🌸 / hostile⚔️ colored system-message banners for random encounters (can be triggered manually from the sidebar).
• Fixed first-message scenario cards overlapping the chat input menu on some themes/resolutions.
• Improved asset auto-tagging detection for given-name mentions such as Patchouli and Flandre.
• Cleaned up non-canon / externally sourced OC entries.

</details>

<details>
<summary><strong>📜 License · References · Thanks</strong></summary>

License — CC BY-NC-SA 4.0

• The creator's original work — system design, code, regex pipeline, Lua scripting, sidebar UI, spell-card cut-in effects, the 文々。新聞 status panel, lorebook structure, asset-prompt design, etc. — is licensed under CC BY-NC-SA 4.0 (Attribution–NonCommercial–ShareAlike). You may remix, adapt, and build upon it as long as you credit Canister and keep it non-commercial, provided that any derivative work is released under the same CC BY-NC-SA 4.0 license.
• Touhou Project original IP (characters, settings, world, music, etc.) belongs to Team Shanghai Alice (ZUN / 上海アリス幻樂団) and follows the official Touhou Project Derivative Work Guidelines. This bot is an unofficial fan work.
• The CC license does not grant rights to the Touhou IP itself; derivative works must separately comply with Touhou guidelines.

© Canister — Unofficial Touhou Project fan work

References & Thanks

• Parts of the parallel-world narrator's operating rules and ideas were referenced and adapted from the "Custom Narrator" lorebook of the Touhou General Module (by its author on the AI-chat community board). Rights to those portions remain with the original author. Thanks to the author for sharing such a solid foundation.
• The initial concept and ideas for this bot were inspired in part by the Touhou Project Simulator bot. Thanks to that work for the inspiration.

</details>