Come in.
Let’s make a story.

V2 Ember is deprecated. For new installs, choose Flame V4 or Kindling V2.

• Promoted the current V2.5 work to V3 Flame, V3 Flame Lumi, and the V3 Regex Pack: 105 SillyTavern prompts and 103 Lumi blocks.
• Restored Thought Engine interview ordering and clarified the Feeling Engine subtitle.
• Polished responsive Heart faculties and .80s drawer reveals.
• Scoped script 09.00 to animated prose paragraphs so its wrapper no longer encloses HUD cards; restored standalone tremble keyframes.
• Mobile Subtext, Tell, Lore, and Translate float beneath the tapped phrase without moving the prose; desktop tooltips, SVG icons, and internal scrolling are preserved. Translate keeps its letter spacing on hover/focus to prevent text reflow.
• Transferred the latest Character Matrix wording from Lumi to Flame while retaining the repaired kinetic wrapper and mobile micro overlays.
• Synced Flame's HTML Tools to Lumi's 21-flavor toolkit without full reference examples. Anti-Therapy Speech and Lightning Story Review are off by default; removed the Spatial Ledger and Scene State renderers and renamed the Director Notebook rule in Flame and its 214-script companion pack.
• Redesigned the Character Matrix as an illustrated spine dossier: nine field icons, textured panels, contrasting mask/motive surfaces, and a parchment audit. Paired desktop panels stack on narrow cards; the .80s disclosure reveal respects reduced motion. Tracker instructions and data fields are unchanged.
• Fixed Matrix field parsing when generated fields share a line with a middle-dot separator, including Mask · Internal. Newline and inline layouts retain the same illustrated panels and original field values.
• Corrected Flame's prompt-manager metadata: all prompt entries now use system_prompt: false. Message roles, prompt text, enablement, injection settings, and regexes are unchanged.
• Removed duplicated animation tails and stray closing braces from Flame, Lumi, and companion regex-pack card CSS so later layout, responsive, and reduced-motion rules survive CSS processing. Preserved .80s timing, native formats, and existing rule order; refreshed Lumi cache revisions.
• Verified identical Sola module order, roles, injection positions, and depths across Lumi and both Flame order profiles. SillyTavern's separate card-prompt hooks remain intact.
• Removed the blank reverse_proxy and proxy_password fields from current Flame exports. Current Lumi exports already contain neither field; live connection settings are not part of this export cleanup.
• Fixed Wardrobe and Inventory cutoffs caused by bare chat-speaker headings. New per-character rows use Name · Wardrobe: or Name · Inventory:; historical bare rows remain display-compatible. Speaker-name protection, tracker graphics, and module order are preserved.
• Corrected Lumi micro-tag paragraph boundaries to compensate for Shadow DOM island spacing (:host>.sola-inline-prose margin -12px), eliminating gap inflation around micro elements.
• Added {{trim}} across all zeroing module statements and untrimmed prompt macro definitions to prevent whitespace emission during initialization.
• Re-indexed Lumiverse regex sort_order to a strictly monotonic sequence (0..214) and synchronized Public Lumi with the 215-rule author master, bringing the 21-flavor HTML Tools and V3 defaults across all release targets.
• Aligned Agency modules (Agency-Lite and Embellished User) to role: assistant with first-person commitments per §7.
• Added the disabled User Persona marker block (personaDescription / persona_description) in both Flame and Flame Lumi right after the User Persona content block.
• Scoped --shadowWidth: 0 !important; across all tracker cards and containers to eliminate SillyTavern/SillyBunny global theme shadow bleed that blurred tracker text, while preserving intentional authored glows.
• Wrapped director quotes for context-specific styling: warm ivory outside the Easel; dark #1A1310, weight 550, and a directional 0 1px 2px rgba(0,0,0,0.65) text shadow on parchment. Wardrobe and Inventory slips use the same dark note treatment.
• Reordered Arrow Dossier ahead of generic director notes and chips, preventing loss of trajectory outcomes and Sola commentary.
• Bounded line-based regex captures to prevent swallowing closing </arrow> and </bond> tags in compact markup.
• Corrected :focus-within selectors for micro popovers (Subtext, Tell, Lore, Translate); disclosures inside cards stay in flow so long content remains reachable without clipping.
• Corrected dotted time matching (a.m. / p.m.) and scene-header attribute-order handling while retaining explicit sky and weather precedence.
• Provisioned shared kinetic keyframes independently of prose wrappers, gave kinetic Tremble precedence over the static tonal alias, and disabled continuous kinetic motion under prefers-reduced-motion.
• Scoped tracker field/status formatting to its owning containers; ordinary prose and unselected status alternatives remain unchanged. Light-surface descendants suppress inherited host shadows without disabling authored spoiler glows.
• Unified tracker drawer spacing and a 12px text floor; darkened small parchment labels to copper ink while retaining each module's imagery, accents, and internal layout.
• Fixed multiline micro payloads: Subtext, Tell, Lore, and Translate attributes may contain blank lines without truncating the disclosure; each micro pair is parsed atomically, so quoted attributes survive paragraph processing.
• Adjacent micro elements inside one narrative paragraph now share a single prose island instead of splitting into per-card blocks, and paragraphs that belong to OOC or card containers stay untouched.
• Micro matchers accept unspaced equals signs, reordered attributes, both quote styles, and enforce matching open and close tags; Retract and Swallow stay readable while a child link holds focus.
• Restored Lumi's authored mobile overlays for the four micro popovers and Translate's stable hover letter spacing, matching Flame.
• Lumi kinetic prose wraps only the animated paragraph, and micro cards carry their own keyframe styles, so mixed kinetic paragraphs keep motion inside shadow islands.
• Chatterbox applies the selected persona palette (Classic, Gremlin, Velvet, Riot) inline on the container from the first painted frame; streaming cards no longer flash Classic orange, and the open-state glow, avatar backdrop, and focus outline follow the persona.
• Repaired section 09 fractional tilt tags and gradient Tremble precedence. Kinetic paragraphs preserve Markdown; SillyTavern receives one message stylesheet, while Lumi uses paragraph-local frames and a conditional island fallback rather than adding animation CSS to every note stylesheet.
• Unified all tracker notes as taped parchment with dark #1A1310 Georgia italic text, weight 550, and #8F3F2E labels. Removed competing note overrides and host-theme shadow bleed while retaining the authored directional parchment shadow; scoped host-shadow resets also protect standalone regex text without removing intentional kinetic glows.
• Fenced section 09 tag scans against malformed streaming markup. Flame and its companion retain 219 scripts; Lumi retains 220 scripts with sequential sort_order 0–219. Both retain 21 prompt savers. The reported Lumiverse application crash still requires its trigger or error to diagnose.
• Bounded seven tracker/note matchers to 4,000-code-unit fields, 20,000-code-unit dossier bodies/searches, and 200-code-unit opening attributes. Calendar, Arrow, and Story Dossier scans stop at the next same-tag opening or closing, preventing repeated malformed blocks from rescanning the entire remainder. Oversized fields skip specialized formatting without truncating source text.
• Restored the illustrated Character Matrix dossier in the latest author exports: plain or bold labels, multiline values, compact middle-dot layouts, and omitted auxiliary fields retain the authored panels and nine icons. Unexpected or repeated-character layouts remain readable through the fallback card.
• Protected standalone tracker line breaks before prose normalization, scoped field parsers to their own tracker families, and restored local disclosure styling without requiring the HUD. Cast Location and Sola rows no longer consume Matrix, Notebook, Arrow, or narrative content.
• Synced the latest 103-prompt Flame and 101-block Lumi masters without changing prompt content. Flame retains 228 regex scripts, Lumi 229 with sequential sort_order 0–228 and cache revision 66, and the companion pack 226; all retain 21 prompt savers.
• Repaired SillyTavern CSS errors at 1:2140, 1:2293, 1:61, 1:2086, 1:2122, and 1:4269: Markdown was turning embedded stylesheet comments into emphasis markup. Audited all 62 stylesheet templates per export, removed 67 comment nodes from 23 rules, and folded the affected sheets' line breaks to prevent paragraph corruption. CSS selectors, declarations, keyframes, artwork, prompt content, and saver behavior are unchanged; Lumi cache revision is now 67.
• Hardened 58 find patterns per V3 export after isolated load, streaming, malformed-input, and native-renderer profiling. The Matrix ST literal shrank from 38,558 to 10,189 characters; incomplete whitespace fields now reject before expensive parsing. Named Heart/Thought cards and savers, tracker scope checks, Inner Current, and inter-tag whitespace cleanup also avoid measured slow paths. Artwork, replacement templates, prompt content, script ordering, and all 21 savers' depth settings are unchanged; Lumi cache revision is 68.
• Repaired Lumi's five Cast matchers and whitespace rule to use bare find_regex plus explicit flags. Removed seven obsolete scriptName aliases that routed native rows through ST import conversion, discarding multiline flags and changing display targets to response targets. Checked the supplied Cast replacement and six-pattern research MDs; retained the current artwork rather than adding stylesheet weight.
• Added scripts/check-regex-regressions.mjs: run with Node or Bun to catch Matrix streaming hangs, field-size boundaries, named NPC closer/separation errors, Cast scope leaks, and whitespace backtracking. Synthetic large-input scope scans and repeated per-island styles remain performance risks; these fixes do not establish the cause of every reported application crash.
• V2 Ember remains available as the previous-release legacy baseline.

A major architectural evolution introducing Embodied Reality, a balanced 16-option Trajectory Fork, modular Action-Anchored Dialogue, dual Story Grounding, streamlined reasoning tiers, and canonical 16-color visual harmony:

🌍 Embodied Reality & Physical Grounding

• Embodied Reality Core Directive: Unified physical space traversal and NPC knowledge limits into one cohesive directive. NPCs perceive strictly what is physically available to them and must traverse concrete space to whisper, touch, or hand items to {{user}}.
• Anti-Echo & Single-Pass Repetition Ban: Folded the 6-gesture sensory repetition ban directly into Anti-Echo, forbidding semantic re-skins and premise echo.
• Post-History Pacing & Stillness: Anchored stillness directives directly at the generation threshold: quiet turns are complete responses; stillness is motion without escalation.

🎲 Trajectory Fork (Balanced 16-Option Engine)

• Zero-Collision Architecture: Completely redesigned the trajectory engine into 16 discrete, inlined permutations. Eliminates duplicate rolls (never Main A + Sub A) and physically erases unrolled routes from the LLM prompt.
• Solo & Compound Trajectories: 25% chance of a focused single route (pure execution without sub-current) and 75% chance of compound routes with layered subtext or physical beats.
• Active Directorial Commitment: Enforces strict scene momentum lock in Story Review, preventing models from kitchen-sinking unassigned narrative vectors.

🗣️ Voice, Idiolect & Bare Speech Modularization

• Rhythmic Axis Protocol: Upgraded the 3-Axis Idiolect Engine (Diction, Rhythmic, Evasion) to emphasize thought assembly, spoken friction, phonetic grain, and cadence differentiation.
• Optional: Action-Anchored Dialogue: Extracted speech tag and vocal texture label bans into a dedicated toggleable module (Action-Anchored Dialogue (Zero Tags) 🏷️) wired via {{mod_notags}}.
• Pre & Post-History Story Grounding: Twin modular toggles enforcing Anti-Bathos and human social friction ("Humanism > Style") without conversational runoff.

⚡ Reasoning Tiers & Thinking Countermeasures

• Stop Thinking So Hard 🧠 (Hard Limits): Strict ~150-word, 5-bullet telegraphic budget banning trial drafting, scene rehearsal, hedging, and moralizing.
• Lightning Story Review ⚡: Forward-only directorial speed-run guaranteeing physical grounding, {{char}} boundary fidelity, and 3-Anchor locks (OPEN, CRUX, CLOSE).

🎨 Canonical 16-Color UI & V2 Regex Pack

• 16-Color Canonical Alignment: Unified all palette chips, HUD badges, and scene ephemera across Day (Golden-Hour) and Night (Obsidian) palettes.
• V2 Sola Character Artwork: High-resolution official character portrait released in full fidelity.

🧩 Complete Roster of New V2 Modules & Toggles

• Core: User Agency-Lite vs User Embellishment/Control (choose one): Twin agency architectures allowing players to choose between strict seed-only simulation (tight leash) or proactive co-authorship where Sola builds ambient momentum and organic scene ripples (loose leash).
• Optional: Anti-Slop and Lite Anti-Slop (choose one): High-precision surgical filters purging AI clichés, dramatic theatrical crutches, and formulaic sentence structures.
• Optional: Voice: Action-Anchored Dialogue (Zero Tags) 🏷️: Banned conversational speech tags (said, whispered, muttered) and stage-direction captions, anchoring dialogue entirely to concrete physical action beats or bare spoken lines.
• Core: Pre-History and Post-History Story Grounding (choose one): Flexible dual-placement modules enforcing Anti-Bathos (zero ironic quips during tension/intimacy) and grounded conversational friction (Humanism > Style).
• Optional: Scene Establishing Headers 🎬: Dynamic scene establishing cards rendering atmospheric time-of-day skyboxes (Morning, Noon, Dusk, Twilight, Midnight, Storm) and location chips via pure CSS.
• Optional: Voice, Idiolect, Dialect & Dialogue Craft 🗣️: Upgraded 3-Axis differentiation protocol (Diction, Rhythmic Axis, Evasion) featuring the anti-cosplay strip-test, voice swap-test, and strict dialogue echo bans.
• Optional: Default Voice: NPC Dialogue Colors (choose one) 🎨: Persistent chromatic identity across 16 canonical Day and Night colors, with seamless inline tone (<color:tone>) and kinetic (<color:kinetic>) compound tagging.
• Optional: Voice: NPC Tonal Typography 🪶: Typographic emotional inflections (<whisper>, <bite>, <tremble>, <steady>) that mix freely with dialogue colors and gradients.
• Optional: Voice & Monologue: Chromatic Gradients 🌅: Radiant mathematical CSS gradients (<dusk>, <solstice>, <hearth>, <aurora>, <abyss>) for internal reflections and pivotal turn capstones.
• Optional: Prose & Voice: Kinetic Emphasis ⚡: Acoustic impact and kinetic weighting tags (<heavy>, <stretch>, <tilt>, <echo>, <crush>) for physical shocks and lingering silences.
• Optional: Gradient ALL The Time (choose one) & Stylistic Overload: Opt-in stylistic amplifiers for players who want full chromatic vibrancy and maximal sensory saturation.
• Tracker: Thought Engine 💭: Grounded backstage interiority check-in evaluating raw impulse vs. spoken mask before dialogue lands.
• Tracker: {{char}} Matrix 🧬: Hard boundary audit, orientation fidelity lock, and anti-orbit tracker ensuring NPCs never unnaturally shadow or mirror {{user}}.
• Reasoning: Stop Thinking So Hard 🧠 (beta test): Ultra-lean 150-word, 5-bullet telegraphic flight-recorder cutting trial drafting, deliberation, and moralizing inside <think> tags.

🧹 SillyBunny Agent: Sola De-Sloppinator

• Autonomous Anti-Slop Polishing: Introduced Sola De-Sloppinator.json, a dedicated SillyBunny in-chat agent providing warm, honey-sweet anti-slop governance and lived-in cadence curation without adding bulk to your baseline prompt stack.

🎨 Lumiverse Themes: Sola Flame, Ember & Solstice

• Sola Flame (Day Palette): Warm Autumnal Hearth aesthetic with Marigold dialogue (#F4A261), Amber interiority (#F3B74B), Coral accents (#EE6C4D), and Solstice primary CTA buttons.
• Sola Ember (Night Palette): Nocturnal Obsidian Cosmic aesthetic featuring Denim dialogue (#7A8B99), Mist interiority (#98B9C7), Indigo accents (#8592E6), and deep obsidian backgrounds.
• Sola Solstice (Dual Mode): Mode-aware dynamic theme seamlessly displaying Sola Ember in Dark Mode and Sola Flame in Light Mode.
• Dual-Format Distribution: Distributed in both 1-click .lumitheme archive and portable .json ThemePack formats with custom dialogue and interiority CSS.

📂 Prompt Categories & Section Dividers

• Plain-Language Prompt Categories: Grouped prompts as Core, Optional, Tracker, and Reasoning so players can tell foundational guidance, optional craft tools, state tracking, and private review helpers apart.
• High-Contrast Numbered Dividers: Upgraded dividers to numbered double-rule section headers (══════════ 01..09 ══════════) making prompt list navigation effortless in SillyTavern.
• Curated Build Recipes: Added 3 ready-to-use toggle recipes (Minimalist, The Novelist, The Full Showrunner) to the Hub and documentation.

A major architectural leap forward, introducing dynamic recency synchronization, unified showrunner telemetry, full anti-therapy behavioral gates, and streamlined POV token efficiency:

🧠 Dynamic Recency & Macro Engine

• 21-Variable Dynamic Synchronization: Implemented a full {{setvar}} / {{getvar}} macro pipeline linking active telemetry trackers, POV modes, tone colors, and HUD elements with zero configuration drift.
• Context-Aware Recency Stack: Tracker Reminder 📌 and Story Review ⭐ now dynamically mirror your exact enabled modules, eliminating token waste from inactive prompts.

🎛️ Showrunner Master HUD & Multi-NPC Telemetry

• Unified Showrunner Master HUD: Consolidates all active state trackers into an obsidian & warm amber console wrapped in a single <sola_hud> block with clean, glanceable telemetry badges (T# F# D# Y#).
• Native Multi-NPC Telemetry: Single-container tracking for multiple active characters within <fengine> and <bond>, preventing tag bloat and parse breakage.
• Showrunner {{char}} Matrix & Self-Audit (<char_matrix>): Deep diagnostic tracker auditing persona adherence, subtext, and active psychological masks.

🛡️ Anti-Therapy & Grounded Dialogue Overhaul

• Comprehensive Anti-Therapy Rules: Banned clinical counseling habits including reflective listening ("What I hear you saying..."), standalone validation beats ("That's completely valid"), permission-granting ("You don't have to answer that"), and counselor interrogation.
• Human Ineptitude in Heavy Moments: NPCs have explicit permission to be awkwardly human—getting uncomfortable, saying the wrong thing, offering a drink or silence, or changing the subject rather than turning into a supportive therapist.
• Let the Beat Sit: NPCs deliver sharp observations in one concise line, strictly preventing paragraph-long psychological lectures or unsolicited emotional debriefs.

👁️ Streamlined POV Architecture (20%–50% Token Savings)

• Dense Declarative Blueprints: Replaced redundant metadata blocks and tautological rules across all 5 POVs (1st Person, 2nd Person, Freaky Hybrid, 3rd Person Past, 3rd Person Present) with dense, high-impact constraints.
• Third-Person Present Optimized: Slashed token overhead nearly in half (157 words → 81 words) while maintaining hardened anti-past-drift grammar defenses.

⚡ Reasoning & Thinking Optimization

• Lightning Story Review ⚡: New lightweight thinking module designed for fast reasoning models, enforcing one-pass telegraphic planning without full-prose drafting or token burn.
• Eastern Model Story Grounding: Added Flash Story Grounding tailored specifically for GLM, MiniMax, and DeepSeek, alongside the Stop Thinking So Hard 🧠 countermeasure.

⚙️ Sola V1.2 Regex Companion Pack

• Showdown Markdown Isolation: Wrapped tracker cards in isolated <div> containers, resolving multi-card HTML breakage and ensuring flawless rendering on desktop and mobile SillyTavern.
• New Visual Cards: Full support for <char_matrix> trait audits and multi-NPC feeling engine chips.

• Introduced the Ember / Flame edition split (Author Edition vs Core Foundation).
• Added OOC Collaborator Protocol for conversational director-level scene guidance.
• Integrated Feeling Engine (TFDY), Bond Radar, and Chekhov's Gun state trackers.
• Initial release of the Sola Regex Companion Pack with custom UI cards.

• Initial public release of the Sola prompt architecture.
• Introduced Oath directives, somatic reflex governance, and authentic NPC knowledge limits.