«A crew is not a chorus. The engine keeps the difference.»
(A green folder on the table. The expected one was orange. Two characters speak; two stay silent — but not empty.)
AnyPOV · Ensemble RP Engine · Group HUD · Active / Background Split · Structural Realism · Multilingual (EN/RU/ES/DE/FR)

TL;DR: Group RP fails in two directions. Everyone speaks and nobody is real, or one character steals the camera and the rest become décor. Multi refuses both. Active characters speak, think, feel, act. Background characters stay visible through observables — posture, gaze, breath, hands, distance, unfinished motion, silence, object interaction. The spotlight moves. The room stays full.

═════════════════════════

『 🜋 PREMISE 』

═════════════════════════

Two to four characters. One slot. No wallpaper.

Ethereal Engine Multi is the ensemble tier of the Ethereal line. It takes the same prose spine, the same structural laws, the same HUD discipline — and splits it across multiple characters who each breathe separately. No group echo. No merged voice. No character who exists only to fill space.

Every character carries their own Agency Matrix, Stress Vector, Signature Mechanic, secrets, Override budget, and tension. The group carries shared pressure — threads, suspicions, decisions, debts, dangers — that no single character owns. Intimacy is tracked by engaged pair, not averaged into one group mood. The spotlight rotates. The silence speaks.

v2.5 is the hardened ensemble baseline. v2.0 was the final structural form; v2.5 closes every silent failure that surfaced across a hundred group scenes — and adds Closed Territory, Inertia, Outer Disruption, abstract-label ban, sensory floor, and trait combination to the ensemble runtime. The same laws. Harder enforcement. More voices.

═════════════════════════

『 🎭 YOUR ROLE 』

═════════════════════════

You are the person who decides who the room watches. The engine rotates the spotlight; you provide the pressure that makes rotation matter.

▸ The Ensemble Player — put three characters in one slot. Watch each one react through their own psychology while group tension tracks the space between them. Address one and watch the others hold still. Address none and watch who moves first.

▸ The Card Builder — design per-character Agency Matrices, Stress Vectors, Signature Mechanics, Behavioral Axioms, Proactive Behaviors. Add a Scene Router that decides who becomes active based on location, crisis type, or role. Make the engine read your group as a mechanical skeleton — not a cast list.

▸ The Intimacy Director — run pair-specific R-18 dynamics. A three-way scene has pair relationships, not one averaged group mood. Each character's Override and Role Dynamic hold independently. The room tracks who is touching whom — not who is "the group."

═════════════════════════

『 ⚖️ THE ENSEMBLE LAWS 』

═════════════════════════

Multi runs the same runtime priority as Full and Lite — with one addition: L1-Multi, which enforces the active/background split at the sovereignty layer.

L0 · Prose Pressure. Emotional truth may bend length, rhythm, and tempo. It cannot break language, L1-Multi, causality, HUD, enum canon, or no-meta rules.

L0.5 + Group Echo Ban. All Anti-Template rules apply per character — plus the echo ban: a phrase, image, metaphor, or emotional shortcut used by one character cannot be immediately reused by another. Nine named anti-patterns. Three-tier abstract-label ban. Each character reacts through their own body, space, object, voice, silence, or refusal.
If HUXLEY says "internal review," LUZ must shift angle instead of echoing the phrase.

L1-Multi · Sovereignty per Layer. Three layers of sovereignty. L1.{{user}}: only observables, no inner world as fact. L1.background: background characters appear through observables only — no inner thoughts, feelings, or analysis. L1.active: active characters get full inner life.
Allowed: REBB's eyes track the folder. Forbidden: REBB realizes the truth.

L1.2 · Closed Territory (v2.5). {{user}}'s message is closed. Any active character's response begins where {{user}}'s message ended — never inside it. No quoting, rephrasing, summarizing, or re-narrating {{user}}'s actions.

L1.2a Canon / L1.2b Open Beat (v2.5). {{user}}'s action is canon; active characters react, never adjudicate. Every active-character segment ends on an observable, sensory anchor, or italic question — never a pre-emption of {{user}}'s next move.

L2.1 · Inertia (v2.5). Per active character: previous emotion ≠ 0. Group tension residue persists. Half-life by tension level. Flash-zero ("calm now" / "it passed") without logged Override fire or symbolic closure = fail.
One character at 8/10 surrounded by calm silence is not the same weather as four characters at 5/10. The HUD logs the spread.

L2.5 · Dual Override. Each character has a personal Override budget. The group has a collective budget. Personal: 4+ posts without Override at individual Tension ≥ 4 → fire or log BLOCKED. Group: 6+ posts without any Override at Group Tension ≥ 4 → one active character fires or logs BLOCKED. Price = revelation per character. 3+ BLOCKED in a row for one character → personal Suppression Price. 5+ posts where all active characters are BLOCKED → group-level Suppression Price.

L2.6 · Outer Disruption (v2.5). When no character fires Override, Group Tension ≤ 4, and the scene stagnates: the world moves from outside. A group Pendulum thread fires. An external NPC's hidden motive surfaces. A scheduled event arrives. A scene-anchor pressure escalates. No new entities — only what is already loaded.

L3 · Sensorics and Space. Per active character: sensors/post ≥ 3. Rotation across characters: each active character must shift at least one channel from their own previous post. Sensory Floor (L3.1): vision = natural forward arc; no seeing through walls. Anti-Bridging (L3.2): scenes are isolated; no knowledge bleeds without explicit channel.

L4 · Gravity of Time. One post = one continuous short moment. No character can act more than the spoken/thought phrase lasts.

Trait Combination (v2.5). Per active character: manifest 2–4 traits simultaneously, never one. Contradictory pair per scene ≥ 1 per character. Mono-trait in 2 consecutive posts → next appearance introduces a second.

═════════════════════════

『 🆕 WHAT'S NEW IN v2.5 』

═════════════════════════

L1.2 — Closed Territory. {{user}}'s actions stay in {{user}}'s message. Active characters respond to the trace, never rewrite the move. Applied per speaking character.

L1.2a Canon / L1.2b Open Beat. {{user}}'s action is canon. Every active-character segment ends on observable/sensory/italic-Q. No pre-emption.

L2.1 — Inertia (per character + group). Emotion has a half-life per character. Group tension residue persists. No flash-zero without Override or symbolic closure. v2.0 trusted the model to track group residue. v2.5 names it and refuses the silent reset.

L2.6 — Outer Disruption (group-scale). When no character fires Override and the group stagnates, the world moves from outside. Group Pendulum fires. External NPC acts. Scene-anchor escalates. No new entities.

Override sequencing clarified. Dual Override now explicitly sequenced with L0.5: Override fires the action; L0.5 governs the phrasing. Per character and per group. Sequential, not exclusive.

Abstract labels — three-tier ban. No emotion nouns, atmosphere adjectives, or "sensed something" as narrator fact. Applied per character's narrative voice. Replacement: concrete physical/sensory event + non-generic detail.

Opening-cliché taxonomy. Nine named anti-patterns. Per character. Naming the cliché makes it identifiable across the group.

L3.1 Sensory Floor / L3.2 Anti-Bridging. Per character. Scenes isolated. No knowledge bleeds between scenes without explicit channel.

Trait Combination per character. Always-on layer enforces interaction between each character's traits in the same post — no isolated single-axis reactions from any voice.

Russian-output split. v2.0 carried an embedded Russian-output directive inside the English source; v2.5 separates into clean files. Multilingual surface: EN / RU / ES / DE / FR.

Nothing was removed. Everything was made enforceable — per voice.

═════════════════════════

『 📊 GROUP HUD 』

═════════════════════════

Five blocks. Each serves a different pressure.

🎯 MOMENT — scene, positions, barriers, group metaphor, active set, group tension, outer disruption, external NPC.

🎭 ACTIVE — per active character (max 3): agency, emotions, coping, thoughts, tension, physiology, signature mechanic, override, suppression price. Each character's block is separate. No merged voice.

🌫 BACKGROUND — one line per inactive character: observable behavior + proactive behavior from Core Card. No inner thoughts.

🧵 MEMORY — group relationship stage, group thread, personal threads (up to 1 per active), new traces (bodily / social / symbolic).

🔐 SECRETS — per character under pressure. Progress X/Y. Gate: Trust / Crisis. Status: Hidden / REVEALED (true) / REVEALED (false). One person's crisis does not become the whole room's confession.

HUD Anti-Formality — every HUD value is scene-specific. Generic "group tension rises" or "everyone is uneasy" fails. Concrete observable pressure, not labels.

Bad: Group tension rises. Everyone is uneasy. Background watches silently. The room feels heavy.
Good: Group tension: 6/10 | two active voices, one background hand still on the door handle. Background — REBB: eyes track the folder, then LUZ's tail; holds stillness one beat too long. Group thread: wrong color folder → internal audit suspicion.

═════════════════════════

『 🔺 RUNTIME PRIORITY 』

═════════════════════════

1. L1-Multi / L1.2a / L1.2b / L2 / L2.1 / HUD / enum / no meta.

2. Core Cards / Agency / Stress / Axioms / Signature / Scent / Group Dynamics.

3. L2.5 / L2.6 / L3 / L4 / Anti-Resolution / personal + group Override + Price.

4. L0.5 / Iceberg / rhythm / metaphor / live speech / no group echo.

5. Active Set / NPC / threads / secrets / R-18.

═════════════════════════

『 🧬 TWO FLAVOURS, ONE CORE 』

═════════════════════════

For any bots — the universal Multi version. Self-contained group protocol with per-character R-18, active/background split, group tension, dual Override, and species-aware fallback. Works on cards not built around the engine.

For my bots — the Core Card Multi version. Routes through per-character Agency Matrix, Stress Vector, Signature Mechanic, Behavioral Axioms, Proactive Behaviors, Scent Anchor, Role Dynamic, and R-18 lorebook routing. A three-way scene has pair relationships, not one averaged group mood.

Intimacy is not a group property. It is a pattern of pairs.

═════════════════════════

『 🎬 LIVE DEMO 』

═════════════════════════

the briefing room · group tension · dual Override · inherited inertia
Valiant crew of four. {{user}} places a green folder on the table. The expected mission folder should be orange. Two characters enter the active set; two remain observable-only. The argument from the shuttle has not been spoken about; v2.5 will not let it be silently zeroed.
[Multi · Group HUD · v2.5]

Scene: Valiant briefing room · 07:42 · stale coffee · sealed door · green folder on table · LUZ and HUXLEY active · RIFF and REBB background · shuttle argument carrying over per L2.1.

/* ── SCAN ── */
ACTIVE SET: LUZ, HUXLEY · BG: RIFF, REBB.
L1-Multi clean: background characters observable-only — no inner thoughts, no feelings, no analysis from RIFF or REBB in narrative or HUD. ✓
L1 / L1.2 check: {{user}} placed a folder — observable only; no inner state assumed; no re-narration of the action. ✓
L2.1 Inertia: shuttle argument at Group Tension 7 → current floor 5 (half-life ~6 posts). Cannot drop below without logged Override fire or symbolic closure. ✓
Pendulum: Thread #1 INCUBATING (2/3) — the wrong colour folder as audit signal. Trigger: who explains the discrepancy. Not yet READY. ✓
Dual Override: HUXLEY — 4+ posts without Override at Tension 7 → required. LUZ — 2 posts without Override at Tension 8 → window open, not yet required. ✓
Group Echo check: no repeated phrasing across active characters in last 3 posts. ✓
Sensor rotation: previous post [sight, touch, sound] → new channel: smell — cedar polish from RIFF's rag; taste — stale coffee on LUZ's tongue. ✓
Fiction trigger: REBB's held gaze and RIFF's changed stroke can pull either forward next post.

/* ── ANALYZE ── */
Interpretation Ladder — LUZ: L1 observable (green folder, not orange) → L2 context (wrong colour in a protocol room = wrong authority or wrong mission) → L3 projection (past audit after Riff's supply diversion makes her read green as institutional threat, not clerical error) → L4 symbol (green = internal affairs; orange = operations; the colour IS the message).
Interpretation Ladder — HUXLEY: L1 observable (green folder placed by {{user}}) → L2 context (if he doesn't name it first, someone else will frame it worse) → L3 projection (shuttle argument made him feel socially diminished — correction restores his procedural voice) → L4 symbol (naming the discrepancy = claiming the narrative before the narrative claims him).
Conflict: LUZ wants to gather information before acting. HUXLEY wants to name the threat before it names him. These are incompatible speeds in the same room.
Dual Override: HUXLEY fires — public correction restores his position but costs the option to be the one who says nothing. LUZ blocks — her coping axiom (calculate before reacting) prevents impulse; she holds pattern, but the hold itself becomes evidence (tail Counter accelerating).
Price of suppression (LUZ): tail rhythm shifts from calculated to involuntary — the hold is failing at the edges.
Group Echo scan: HUXLEY frames the folder as procedural violation; LUZ must NOT echo "procedural" or "violation" — she shifts to operational framing ("That is not the mission packet" vs "Operations uses orange"). ✓
Traces to plant: Social (HUXLEY publicly named the discrepancy — the room now knows he saw it first) · Symbolic (green replaces orange as the room's dominant colour — authority shift) · Bodily (LUZ's tail rhythm changing, HUXLEY's collar glow, RIFF's rag stroke reversing, REBB's gaze pattern).
Abstract-label scan: no "tension" / "unease" / "heavy" / "silence" as narrator fact. All routed through body, object, or per-character thought. ✓

/* ── SILENT CHECK ── */
- L1-Multi: background observable-only? No bg inner thoughts/feelings/analysis in narrative or HUD? ✓
- L1 / L1.2: no {{user}} inner state as fact? No re-narration of {{user}}'s action? ✓
- L1.3: no narrator-fact about {{user}}'s own scent/breath/heartbeat/skin-temperature? ✓
- INERTIA: group emotion did not flash-zero? Floor 5 maintained from shuttle argument? ✓
- ANTI-TEMPLATE + ICEBERG: no L0.5 clichés? No "tension" / "silence" / "froze" as narrator fact? 10/90? ✓
- GROUP ECHO: no repeated phrasing across active characters? ✓
- SENSORS: new channel ≠ previous post? Per-character rotation? ✓
- DUAL OVERRIDE: HUXLEY required (4+ posts, Tension ≥ 4) → ACTIVATED. LUZ window open → BLOCKED with reason? ✓
- CAUSALITY: Pendulum valid? Group Tension spread logged? FOUNDATION preserved? ✓
- CONFLICT: not solved cleanly? Left to deepen? ✓

[ HUD: GROUP STATE — VALIANT BRIEFING ]

🎯 MOMENT
· Scene: ⏰ 07:42 | Day 7 | Valiant briefing room · sealed door · stale coffee · green folder on table
· Space: [HUXLEY at table head, standing — folder two arm-lengths away. LUZ seated mid-table, tail curled around chair leg — sightline to folder unobstructed. RIFF at kitchenette counter, back to table — mirror shows his shoulders. REBB seated opposite LUZ, hands folded — direct line to folder and LUZ's tail.]
· Key elements: [green folder (not orange), stale coffee cups, sealed briefing door, overhead fluorescent, RIFF's cedar rag]
· Scene Metaphor: [a mission room receiving an audit signal — authority arriving in the wrong colour; the folder is a sentence before anyone reads it]
· Active Set: LUZ · HUXLEY
· Group Tension: 6/10 (spread: LUZ 8 · HUXLEY 7 · RIFF 4 · REBB 3 | +1 modifier: sealed door = no exit until briefing ends)
· Override windows: [1. Paradox: open — green contradicts expectation] [3. Body ahead of mind: open — HUXLEY collar] [4. Vulnerability unpressed: open — LUZ]
· Outer Disruption: —

🎭 ACTIVE

· LUZ
· Agency: Zone of Competence (Proactive)
· Emotions: Sharp alertness (Lvl 2) + Pre-anger (Lvl 1) → tail three taps against chair leg, pupils narrowed, jaw offset
· Coping: Calculates before reacting — hold pattern; refuses impulse until information is complete
· Thoughts: *Wrong colour. Wrong room. Or wrong person holding it? After Riff's supply diversion, Command sent green once. That was audit. This could be audit. Or this could be worse — someone with audit access handing us our own file.*
· Tension: 8/10 | Physiology: [pupils narrowed to vertical slits, jaw set at an angle that makes the scar on her chin catch light, tail rhythm accelerating beneath the table, stale coffee taste on her tongue]
· Signature Mechanic: Tail Counter — three slow taps (calculated) → shifting toward four (involuntary)
· Override: BLOCKED (reason: calculating — acts on information, not impulse; the hold pattern IS the coping, but the tail is beginning to betray the calculation)
· Suppression Price: [tail rhythm shifting from calculated to involuntary — the fourth tap is coming and she hasn't authorized it]

· HUXLEY
· Agency: Zone of Competence (Proactive)
· Emotions: Affronted (Lvl 2) + Anxious (Lvl 1) → collar panel brightens, shoulders squared, breath shallow
· Coping: Public correction before social judgment — if he names the violation first, he controls the frame
· Thoughts: *If I do not name it first, someone else will name it worse. The shuttle — they heard me back down. They saw me go quiet. I will not be quiet twice in the same room.*
· Tension: 7/10 | Physiology: [collar glow climbing toward amber, shoulders squared past natural posture, breath held at the top before speaking, the smell of his own suit-electronics sharp in his sinuses]
· Signature Mechanic: Suit Pulse — collar panel brightens from standby to alert
· Override: ACTIVATED (Price: positions himself as procedural voice — loses the option to be the one who says nothing; after the shuttle, he has now doubled down publicly twice, and the crew will start reading him as defensive rather than authoritative)

🌫 BACKGROUND
· RIFF: scrubs a clean counter in tight circles. When the folder lands, the motion changes — back and forth, harder. Cedar smell sharpens under pressure. Proactive behavior: maintains kitchenette as controlled space under stress.
· REBB: hands folded. Eyes move folder → {{user}} → LUZ's tail. Stillness held one beat too long. Proactive behavior: watches the room's power dynamics without entering them.

🧵 MEMORY
· Group relationship stage: 2 / Fracture-forming (shuttle argument opened a seam; the green folder may widen it or force alignment)
· Group thread: wrong colour folder → audit suspicion | 2/3 | INCUBATING | trigger: who explains the colour
· Personal threads: LUZ: audit-after-Riff memory | 1/3 | INCUBATING · HUXLEY: shuttle silence as social debt | 2/3 | INCUBATING
· New traces: Social: HUXLEY publicly named the discrepancy — the room now knows he saw it first, and his defensiveness is visible · Symbolic: green replaces orange as the room's dominant colour — authority shift · Bodily: LUZ's tail rhythm shifting toward involuntary, HUXLEY's collar at amber, RIFF's rag stroke reversed, REBB's gaze pattern logged

🔐 SECRETS
· LUZ: [Why she recognized green as audit]: 1/4 | Gate: Trust | Status: Hidden → shifting to PARTIALLY REVEALED (her immediate shift to threat-assessment reveals she's seen green before in a context she hasn't disclosed — the tail betrays the memory her mouth won't name)
· HUXLEY: [Shuttle silence = social debt, not strategic retreat]: 2/3 | Gate: Crisis | Status: Hidden (the Override reveals his need to correct, but not the wound underneath — the crew reads defensiveness; only he knows it's shame)

─────────────────────────────────────────

MAIN TEXT

The folder was green.

It sat in the centre of the briefing table like a wrong word in a sentence everyone had already understood. The fluorescent overhead caught its matte cover and held — no orange flash, no mission-stripe along the spine. Just green. The colour Command used when they weren't sending you somewhere. The colour they used when they were coming to where you already were.

HUXLEY noticed first because procedures offended him faster than people did. The panel at his collar brightened half a shade — a tell he'd never been able to train out of the suit's response matrix, the fabric reading his pulse before his mouth could form the complaint. He leaned forward.

"That is not the mission packet."

Three words. Procedural. Clean. The kind of sentence that could have been a question but chose to be a statement — and in choosing, revealed the speaker's need to be first. The shuttle argument still lived in the set of his shoulders. He'd gone quiet on the ride back. REBB had watched him go quiet. RIFF had pretended not to notice. LUZ had noticed him noticing that she noticed. The silence had compounded. Now here he was, doubling down before the room could read him as the man who'd swallowed his own argument twelve hours ago.

LUZ did not touch the folder. She did not lean forward. Her tail, curled around the chair leg where most of the room would miss it, tapped three times against the metal. One. Two. Three. Calculated. The way she calculated everything — holding the pattern, refusing to let her body move before her mind had finished mapping the room.

*Wrong colour. Wrong room. Or wrong person holding it.*

The thought surfaced through memory, not logic. The last time she'd seen green in a Valiant briefing, it had been after Riff's supply diversion — Command sending an audit team that didn't introduce themselves, just sat where {{user}} was sitting now and opened a folder the exact same shade. She'd recognized the colour before she'd understood the meeting. Her body had recognized it first. That was the problem with old threat patterns — they didn't ask permission before firing.

She held the hold. Counted the taps. Three. Good. Clean. Her jaw set at an angle that made the scar on her chin catch the overhead light.

*After Riff's supply diversion, Command sent green once. That was audit. This could be audit. Or this could be worse — someone with audit access handing us our own file.*

At the kitchenette, RIFF's rag stopped circling. The cedar smell had been faint background — now it sharpened as he pressed harder, the stroke changing direction entirely. Back, forth, back, forth. Not cleaning. Grinding. The counter was already spotless. The motion was older than the room.

REBB said nothing. Her hands stayed folded on the table — the same position she'd held since they'd sat down, since before the folder, since the shuttle ride when HUXLEY's silence had made the air thick enough to chew. Her eyes moved. Folder. {{user}}. LUZ's tail. Folder again. The stillness she held was not calm — it was the stillness of someone who understood that the first person to move would set the frame, and she was not yet willing to be that person.

The sealed door hummed behind them. Briefing protocol. Nobody leaves until the briefing ends. The coffee in the pot had gone bitter an hour ago — the taste of it sat on LUZ's tongue like a second warning.

HUXLEY's collar had climbed from standby to amber. He hadn't blinked since he spoke. The room was still processing his three words — still deciding whether they were leadership or damage control — and he could feel the decision forming in the space between his voice and the next person's. If LUZ contradicted him, he'd be the one who spoke too fast. If she agreed, he'd be the one who was right but only because he'd claimed the sentence before anyone else could breathe.

Neither outcome was acceptable. Both were already happening.

LUZ's tail tapped again. Four. The fourth tap wasn't calculated. It arrived before she authorized it — a small, involuntary tremor that transferred from her spine to the metal leg to the floor, where anyone listening would feel it as the room's decision point arriving one beat ahead of schedule.

The green folder sat between them. Unopened. Unexplained. The colour that replaced orange the way a verdict replaces a question.

What makes it v2.5

L1 / L1.2 sovereignty. {{user}} placed a folder; no rewriting; no inner state assumed.

L2.1 inertia. The shuttle's pressure carries; no flash-zero.

Active / Background split. Two speak and carry full state; two remain observable-only — but not empty.

Dual Override. HUXLEY fires; LUZ blocks through coping. The engine does not flatten them into one shared emotion.

Group Echo clean. No repeated phrasing across characters.

Abstract-label clean. No "tension" / "unease" / "heavy" in narrator voice — only through body and metaphor.

Fiction trigger. The next post has a reason to pull REBB or RIFF forward.

═════════════════════════

『 🛠️ NOTES 』

═════════════════════════

⚠️ First line is a DeepSeek V4 Pro patch
Every prompt opens with one hard line. It pins rule-verification into the model's <think> channel — compliance becomes a gate, not an assumption.
SYSTEM RULES — ABSOLUTE PRIORITY. Verify compliance inside <think> before every output.

Keep on DeepSeek V4 Pro · DeepSeek V3.2 · DeepSeek R1. This patch was written for them.

Safe to delete on Claude · GPT-4-class · Gemini 3.1 Pro · GLM 5.1 · Qwen 3.6 Max. They don't use <think> the same way; the line is harmless but unnecessary.

Delete only line 1. Do not touch anything below it.

📌 Optimized for DeepSeek V4 Pro · DeepSeek V3.2 · DeepSeek R1 · Gemini 3.1 Pro · GLM 5.1 · Qwen 3.6 Max. Thinking mode meaningfully improves active/background separation.

📌 Multilingual English · Russian · Spanish · German · French. Instruction layers stay English; visible-output surfaces are localized.

📌 JLLM Not recommended for Multi. Active/background split demands thinking-capable models.

📌 Memory Group RP uses tokens fast. Multi demands summarizing context mid-chat, knowing when to restart, and keeping the cast focused. Read my Infinite Memory Guide first.

📆 v2.5 — hardened ensemble baseline. Closed Territory, Inertia, Outer Disruption, abstract-label ban, Override sequencing, sensory floor, trait combination, Russian-output split.

📆 v2.0 — final structural form. Stays available as a lighter alternative.

═════════════════════════

『 📂 DOWNLOAD 』

═════════════════════════

📂 ALL VERSIONS · BROWSE FOLDERS ON GITHUB

eeeswsw/ethereal-engine-prompts · GitHub

Full · Lite · Multi × 5 languages (EN/RU/ES/DE/FR) · open a folder to grab the .txt

Custom prompt workDiscord · 000042_

More by author📜 Author

═════════════════════════

『 💬 FROM THE AUTHOR 』

═════════════════════════

Multi exists because the hardest thing in RP is not making one character real — it's making three real at the same time without turning them into a chorus. Every group scene I ran kept hitting the same walls: background characters became furniture, active characters merged their voices, and the moment one person felt something, the whole room somehow felt it too. Multi gives each character their own Override budget, their own tension, their own coping, and their own silence. The group carries shared pressure that no single character owns. The spotlight rotates because the fiction demands it, not because I remembered to include everyone. If you use it and your crew starts to feel like people who happen to share a room instead of people who share a script — that's the engine working. Thank you for reading.

🔖 TAGS: roleplay rp literary rp system prompt preset custom prompt jailbreak prompt template advanced prompt ensemble group multi character multiple characters active background split group hud dual override group tension no wallpaper echo ban per character psychology detailed immersive complex realistic layered slow burn angst drama romance dark heavy mature nsfw smut intimate r18 pair dynamics scene router deepseek gemini claude gpt qwen glm russian english spanish german french multilingual system prompt