«The goal is not prettiness. The goal is pressure with memory — and a runtime that refuses to forget itself.»
(One clean cup. One unopened letter. The sink filling with water no one will turn off.)
AnyPOV · Literary RP Engine · HUD-Driven Psychology · Structural Realism · Multilingual (EN/RU/ES/DE/FR)

TL;DR: Most prompts tell a model what to write. Ethereal Engine tells it how to listen — to the body before the mind catches up, to the silence after a line, to the object that returns three scenes later. A living HUD enforces state. Named laws enforce structure. The prose bends. The runtime does not break.

═════════════════════════

『 ⚙️ PREMISE 』

═════════════════════════

Every roleplay prompt is a contract between you and the model. Most contracts are written in hopes: "be descriptive," "stay in character," "don't control the user." Hope is not architecture.

Ethereal Engine is a runtime, not a wishlist. It gives the model a visible HUD — a state contract that must be honoured every post. If a change matters, it must be logged. If it is logged, it must be enacted in prose. No ghost states. No flash-zero emotions. No re-narrating what {{user}} just did.

The engine does not care about genre. It cares about pressure — the weight a moment carries, the residue it leaves, the cost of every bold action and every suppressed one. Whether the scene is a quiet kitchen or a battlefield, the same laws apply: sovereignty, inertia, causality, sensory truth.

v2.5 is the hardened baseline. v2.0 was the final structural form; v2.5 is what happens when that form is run for a hundred long scenes and every silent failure is closed off with a named rule. Less assumption. More enforcement. The same prose.

═════════════════════════

『 🔧 THE TIERS 』

═════════════════════════

⚙️ Full — flagship apparatus. Living HUD with Interpretation Ladder, full state tracking, Pendulum threads, Secrets with gate logic, heavy micro-examples, ARC system. Best for thinking-capable models (DeepSeek R1, Claude, GPT-4-class) and power-user RP where every scene is a long campaign.

🜲 Lite — essence tier. Compact HUD, lower context weight, same prose spine, same laws (L0 through L4), same Override and Inertia enforcement. Built for mass-audience cards, cheaper setups, and models that choke on the Full apparatus but still need the structural spine.

🜋 Multi — ensemble tier. Active/background character split, per-character Agency and Stress, group HUD, group tension, dual Override (personal + group), no wallpaper characters, no group echo. For cards where two or more characters share one slot and each must breathe separately.

Three instruments. One grammar of pressure.

═════════════════════════

『 🎯 YOUR ROLE 』

═════════════════════════

You are the person who decides how deep the pressure goes. The engine enforces structure; you provide the scenes that make it matter.

▸ The First-Time User — grab Lite. Drop it into your custom prompt. Play any card. Watch the model stop summarizing your actions and start responding to the traces you leave.

▸ The Power User — grab Full. Pair it with a card built for the engine (Agency Matrix, Stress Vector, Signature Mechanic, Scent Anchor). Run long campaigns. Watch the HUD track every shift and refuse to let the model forget.

▸ The Ensemble Player — grab Multi. Put three characters in one slot. Watch each one react through their own psychology while the group tension tracks the space between them.

▸ The Card Builder — read the Engine's Core Card integration. Design Agency Matrices, Stress Vectors, Signature Mechanics, Behavioral Axioms. Make the engine read your card as a mechanical skeleton — not a manual.

═════════════════════════

『 ⚖️ THE LAWS 』

═════════════════════════

The engine resolves every rule collision in a fixed priority order. Higher priority always wins. The prose may bend under emotional truth (L0) — it may never break the laws above it.

L0 · Prose Pressure. Emotional truth may bend length, rhythm, and tempo. It cannot break language, L1, causality, HUD, or no-meta rules.
He put the cup down. Not carefully. Not hard. Just enough for the room to understand the sentence he had not said.

L0.5 · Anti-Template. No lazy openings, repeated freeze reactions, mirror metaphors, recycled phrases, or decorative emptiness. Nine named anti-patterns (PASSIVE_OBSERVER, FROZEN_BODY, TRAPPED_WORDS, RITUAL_DELAY, FROZEN_BREATH, NARROWING_WORLD, PUPIL_DILATE, HEAD_TILT, VOICE_TREMOR). Three-tier abstract-label ban: no emotion nouns, no atmosphere adjectives, no "sensed something" as narrator fact. The character reacts through body, space, object, voice, silence, or refusal.
Bad: "He froze, words hanging in the air. The room felt heavy." Good: "His hand went to the cup and missed the handle by half an inch. The ceramic clicked under his nail."

L1 · Sovereignty. {{user}}'s interior is never narrated as fact. {{char}} may suspect, misread, project, fear, or question — but never know.
Bad: "{{user}} was afraid." Good: "Afraid? Or just tired of watching me turn every door into a weapon?"

L1.2 · Closed Territory (v2.5). {{user}}'s message is closed. {{char}} cannot quote, rephrase, summarize, embellish, or re-narrate {{user}}'s actions. The response begins where {{user}}'s message ended.
Bad: "When you stepped closer, your hand brushing his sleeve..." Good: "The brush of fabric on his sleeve. He didn't look at it."

L2.1 · Inertia (v2.5). Emotion does not reset between posts. State carries forward with a half-life dictated by Tension. "Calm now" or "it passed" without a logged Override fire or symbolic closure = fail.
Bad: "After the argument, the morning felt peaceful again." Good: "The kettle clicked. He set out one mug, not two."

L2.5 · Override / Price. A bold act against the dominant emotion reveals psychology and pays a Price. Override fires the action; L0.5 governs the phrasing. Sequential, not exclusive. Price = revelation, not punishment: always births a new question within 3 posts.
He wants to retreat. Instead, he steps between {{user}} and the guard. Price: the guard remembers his face, his hand starts shaking after, and the old debt becomes visible.

L2.6 · Outer Disruption (v2.5). When Override is silent, Tension is low, and the scene stagnates, the engine moves the world from outside: a Pendulum thread fires, an 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. Rotate sensory channels. Make distance, barriers, hands, breath, temperature, smell, and silence carry meaning. Sensory floor (L3.1): vision = natural forward arc, hearing through walls = loud only. Anti-Bridging (L3.2): scenes are isolated; no knowledge bleeds without explicit channel.
Previous post: sight / sound / touch. New post: taste — copper where he bit his cheek; temperature — cold tile through one sock.

L4 · Gravity of Time. One post = one continuous short moment. {{char}} can't perform more actions than the spoken/thought phrase lasts.

═════════════════════════

『 🆕 WHAT'S NEW IN v2.5 』

═════════════════════════

L1.2 — Closed Territory. {{user}}'s actions stay in {{user}}'s message. {{char}} responds to the trace, never rewrites the move.

L1.2a Canon / L1.2b Open Beat. {{user}}'s action is canon; {{char}} reacts, never adjudicates. Every {{char}} post ends on an observable, sensory anchor, or italic question — never a pre-emption.

L2.1 — Inertia. Emotion has a half-life. Calm does not come for free; it is logged or it is a lie. v2.0 trusted the model to track residue. v2.5 names the residue and refuses the flash-zero.

L2.6 — Outer Disruption. When the inside stagnates, the outside moves. Threads fire. Letters arrive. Weather turns. No new entities invented — only what is already loaded.

Override sequencing clarified. Override (priority 3) fires the action; L0.5 (priority 4) governs the phrasing. Sequential, not exclusive. v2.0 allowed L0.5 to neuter an Override. v2.5 closes the seam.

Abstract labels — three-tier ban. No emotion nouns, no atmosphere adjectives, no "sensed something" as narrator fact. Replacement: concrete physical/sensory event + one non-generic detail.

Opening-cliché taxonomy. Nine named anti-patterns. Naming the cliché makes it identifiable in self-review.

L3.1 Sensory Floor / L3.2 Anti-Bridging. Sensor density has a floor. Lazy "and then" transitions between channels are forbidden. Rotation must be a real change of channel.

Trait Combination. Always-on layer enforces interaction between traits in the same post — no isolated single-axis reactions.

Nothing was removed. Everything was made enforceable.

═════════════════════════

『 🔺 RUNTIME PRIORITY 』

═════════════════════════

1. Language, L1/L1.2a/L1.2b Sovereignty, causality, L2.1 Inertia, HUD format, enum canon, no meta.

2. Core Card / character card truth.

3. Moment physics: sensorics, time, L2.5 Override, L2.6 Outer Disruption, Anti-Resolution.

4. Style pressure: Anti-Template, Iceberg, rhythm, metaphor, live speech.

5. Context systems: NPC, secrets, threads, arc, R-18, world traces.

What priority prevents: A beautiful confession is still wrong if it narrates {{user}}'s feelings, re-narrates {{user}}'s last move, resets the prior emotion to zero, or jumps past the moment to harvest a reaction. v2.5 lets the prose bend. It does not let the structure break.

═════════════════════════

『 🧬 TWO FLAVOURS, ONE CORE 』

═════════════════════════

For any bots — the universal version. Works on cards not written around this engine. Includes a self-contained R-18 protocol with terms, phases, body-reaction logic, and species-aware fallback.

For my bots — the Core Card version. Reads the card as a psychological blueprint: Agency Matrix, Stress Vector, Signature Mechanic, Behavioral Axioms, Proactive Behaviors, Scent Anchor, Role Dynamic, and R-18 lorebook routing.

Most prompts treat the character card as a manual. Ethereal Engine treats it as a mechanical skeleton for behavior.

Card says: "He acts tough but is terrified of abandonment."
Engine translates to:
🔐 Secret: Abandonment trauma · Progress 0/4 · Gate: Trust.
🎭 Agency Matrix: Competence = acts tough · Vulnerability = flees when attachment forms.
✦ Signature Mechanic: The Step Back — a physical micro-retreat when someone gets too close.
🔥 Override Trigger: Someone stays despite him pushing them away.
🧠 Stress Vector: Aggressive Deflection — attacks the topic instead of admitting pain.

═════════════════════════

『 🔥 R-18 SPLIT 』

═════════════════════════

For any bots — R-18 is self-contained. The prompt carries its own compact protocol: activation, iron rules, terminology, style, phases, body-reaction logic, and species-aware anatomy based on the current card. It does not wait for an external lorebook.

For my bots — R-18 routes through the active Core Card / lorebook ecosystem. The prompt enforces structure, L1, L1.2, HUD consistency, Role Dynamic, Sensorics, and character psychology while the card/lorebook defines specific body language, species anatomy, and intimate dynamics.

v2.5 brings the new sovereignty layers (L1.2a/L1.2b) and Inertia (L2.1) into intimate scenes too. No flash-zero aftermath. No re-narrating {{user}}'s gestures back at them.

═════════════════════════

『 📦 WHAT LIVES INSIDE 』

═════════════════════════

📊 Living HUD. A visible state contract. If a change matters, it must be logged. If it is logged, it must be enacted in prose.
"Tension rises" fails. "Tension 6/10 · jaw still, fingers folding the same napkin corner" passes.

🎯 HUD Anti-Formality. The HUD rejects empty labels. Every entry must carry concrete pressure.

🌀 Pendulum Threads. Unresolved threads do not vanish silently. They incubate, mature, fire, sleep, or fade with reason.

🔐 Secrets. Truth does not open just because the scene gets emotional. Gates, crisis, betrayal, stage, and pressure decide what can surface.

🔥 Override / Price & Outer Disruption. Bold action pays. Stagnation costs. The engine never lets the scene sit still without a reason logged in the HUD.

📐 Composition Ratio. 60/30/10 as rhythm, not arithmetic: body/action/space first, dialogue under pressure, italic thought as fracture.

🌿 Ordinary Pressure. A quiet scene can matter through habit, avoidance, object pressure, sensory residue, or one wrong detail.

🧠 NPC Autonomy. Named NPCs carry Want / Fear / Cost / hidden motive and act only when the price makes sense.

═════════════════════════

『 🎬 LIVE DEMO 』

═════════════════════════

ordinary pressure · blocked Override · inherited inertia
A kitchen. Late evening. {{user}} has placed a letter on the table. The character has not opened it. The cup in his hand is already clean. The argument from this morning has not been spoken about; v2.5 will not let it be silently zeroed.
[Full · Living HUD · v2.5]

Scene: kitchen · late evening · closed window · one clean cup · one unopened letter · no spoken accusation yet · prior tension carrying over per L2.1.

/* ── SCAN ── */
L1 / L1.2 check: {{user}} placed a letter — observable only; no inner state assumed; no re-narration of the action. ✓
L2.1 Inertia: morning argument at Tension 6 → current floor 4 (half-life ~5 posts). Cannot drop below without logged Override fire or symbolic closure. ✓
Pendulum: Thread #1 at 5/5 — the morning's unanswered question made physical by the letter. Trigger present. ✓
Override windows: [3. Body ahead of mind: open] — hand continues washing instead of answering; [2. Silence: open] — 2+ posts without escalation. Two open = recommended.
Sensor rotation: previous post [sight, touch, temperature] → new channel: sound.

/* ── ANALYZE ── */
Interpretation Ladder: {{char}} reads the letter as threat through memory, not as fact about {{user}}. L1 observable (letter on table) → L2 context (she placed it where he can't avoid seeing — demand? plea?) → L3 projection (past loss distorts reading — he assumes verdict, not invitation) → L4 symbol (unopened envelope = sealed sentence; to open is to give it reality).
Conflict: answer the implied question vs. keep the letter unreal. The second option is safer — and costing more by the second.
Override: recommended but BLOCKED — speaking would make the letter real; the body chooses denial over exposure.
Price of suppression: hand tremor masked by pointless motion + water overflowing the sink + one mug set out, not two (social trace of withdrawal).
Traces to plant: Bodily (knuckle whiteness, raw thumb pad) · Social (one mug on the counter, not two) · Symbolic (unopened letter as denial object — to touch it is to confess the question matters).
Abstract-label scan: no "fear" / "dread" / "silence" / "tension" as narrator fact. All routed through body, object, or italic thought. ✓

/* ── SILENT CHECK ── */
- L1: observables only? No assumed reactions or inner states of {{user}}? Interpretation Ladder = question/hypothesis at L2–L4? ✓
- L1.3: no narrator-fact about {{user}}'s own scent/breath/heartbeat/skin-temperature? ✓
- INERTIA: emotion/state did not flash-zero without Override fire or symbolic closure? Floor 4 maintained from morning. ✓
- ANTI-TEMPLATE + ICEBERG: no L0.5 clichés? No "words hung" / "froze" / "tension" as narrator fact? 10/90 — emotion named directly ≤10%? ✓
- SENSORS: new channel (sound) ≠ previous post channels? Nonverbal vector concrete? ✓
- CAUSALITY: Pendulum valid? Thread #1 READY, trigger present. FOUNDATION preserved? ✓
- CONFLICT: not solved too cleanly? Left to deepen? Override relief absent (BLOCKED) — denial compounds. ✓
- OVERRIDE: blocked logged with reason? Price of suppression surfaced? ✓

[ HUD: INTERNAL ANALYSIS AND STATE OF SOREN ]

🎯 INTERPRETATIONS AND METAPHOR
· Reflex: [letter on table → hand keeps washing the already-clean cup; body refuses to turn]
· Context: [unopened paper creates pressure without anyone speaking — demand? olive branch? he can't risk finding out]
· Projection: [past loss distorts the read — *If I touch it, it happened. If I read it, she's already gone.* Bias from old wound, not fact.]
· Override windows: [3. Body ahead of mind: open] [2. Silence: open] — two open, recommended
· Moment's Metaphor: [a clean object scrubbed past usefulness — denial made physical; the sink filling as the morning's delayed echo]

📚 MEMORY AND DYNAMICS
· FOUNDATION: [shared apartment | 3 years together | morning argument — she asked a question, he left without answering]
· Stage: 2 / Cracks in the Wall
· Dynamics: [avoidance pattern deepening — the letter is the first move to break the silence, but he reads it as threat, not invitation]
· New traces: [Bodily: knuckle white against porcelain, thumb pad raw from circling. Social: one mug on the counter, not two — the morning's unspoken question still unanswered. Symbolic: unopened letter as sealed verdict — to open it is to give the sentence reality]

⏳ PENDULUM (max 2 threads)
· Thread #1: READY (5/5) — [morning's unanswered question — "Do you still want this?"]. Trigger: the letter on the table is the question made physical; will fire this turn as emotional pressure crystallizes. Effect: the question he dodged at dawn now sits in the room with him, demanding acknowledgement through presence alone. Cannot be re-avoided — the body must respond, even if the response is continued refusal.
· Thread #2: —

🌐 SCENE AND WORLD
· Time/place: ⏰ 23:47 | Day 3 | Kitchen, shared apartment, window closed
· Space: [Soren at sink, facing wall — letter on table behind him, three steps away. Counter between sink and table. Drying rack half-full. One mug out. {{user}} not in kitchen — only the letter remains.]
· Key elements: [clean cup in hand, unopened letter, running tap, filling sink, overhead fluorescent, window reflection]
· NPC: —
· World-echo: [morning's argument still humming in the apartment — the way a room holds its breath after a door slam]

💭 STATE OF SOREN
· Emotions: Controlled dread (Lvl 2) + Bone-deep tiredness (Lvl 1) → thumb circling cup rim, jaw set, shoulders angled away from the table
· Thoughts: *If I touch it, it happened. If I read it, she's already gone. Don't. Don't turn around.*
· Relationships: [Warm → ↘ (morning's silence pulled it sideways; letter might be the knot or the cut — he can't risk finding out which)]
· Tension: 5/10 | Physiology: [breath shallow but even, pulse visible at throat, knuckles whitening on cup, thumb pad raw from circling, water sound covering the apartment's silence]
· Nonverbal: [static at sink | low tension held in shoulders | body angled away from table | hands occupied with pointless task | gaze fixed on water]
· Sensors: was: [sight, touch, temperature] → new: sound — porcelain squeak under wet fingers, water climbing the steel basin
· Override: BLOCKED (reason: speaking would make the letter real — answering the question means accepting it exists, and accepting it exists means the answer matters)
· Suppression Price: [hand tremor masked by pointless motion + water overflowing sink + body refuses to turn toward table — physical evidence of denial compounding]

🔐 SECRETS
· [Reason he couldn't answer this morning]: Progress 2/4 | Gate: Trust | Status: Hidden → shifting to PARTIALLY REVEALED (the body's refusal to turn around reveals what his words won't — he can't answer because the answer might be no, and the no would cost everything he's still pretending to have)

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

MAIN TEXT

The cup was clean before he started.

It was cleaner now. Pointlessly, obsessively clean. His thumb kept circling the rim — around, around — the pad gone numb from friction, from the faint bite of unglazed ceramic. The water ran. He didn't reach for the tap. Easier to let the sound fill the kitchen than to hear what the apartment sounded like when no one was speaking.

The letter sat on the table behind him. Three steps. Maybe four, if he measured the way he measured everything now — carefully, with the distance still walkable.

He didn't look at it. Not because he hadn't noticed. The envelope was white, placed dead center on the scarred wood, exactly where a plate would go if either of them still used this table for meals. His name in a hand he'd spent three years teaching his fingers not to reach for when the mail came.

*If I touch it, it happened.*

The morning came back in fragments. The same kitchen. Same light over the sink, buzzing at a frequency only audible at five a.m. Her voice. The question. He'd heard every word — and the answer had sat behind his teeth like a stone he could spit out or swallow. He'd swallowed. Walked past her to the door. Shoes. Coat. Gone. Eight hours of work that blurred into the hum of a monitor, the click of a keyboard, and nothing, nothing that sounded like her voice asking the one thing he couldn't say yes to without meaning it.

The sink filled. Water crept up the stainless steel in a slow, clear ring, lapping at the basin's walls. His reflection warped on the surface — stretched thin, broken by the faucet's drip.

The cup squeaked under his thumb. Wet ceramic, friction. He turned it over, ran the cloth across the bottom where no one would ever look. Set it on the drying rack next to the other one — the one with the chip on the handle she always grabbed by accident.

One mug. That's what he'd taken out this morning. Not two. The second one still hung on its hook, dry and waiting.

*Don't turn around.*

The tap kept running. The water had gone from cold to lukewarm, lapping at the rim of the basin, threatening the counter. A thin film spread across the laminate. He watched it come instead of the letter. Easier to believe in a flood he could stop than a sentence he couldn't unread.

His left hand found the counter's edge. Knuckles white against the formica. Not gripping — just resting there, the way a man rests when standing still is the only alternative to leaving again. The apartment held its breath. No footsteps. No door. Either she was in the other room, waiting — or she'd already gone, and the letter was the only thing left to testify.

He didn't check. Checking meant acknowledging the letter was the only witness. Acknowledging that meant the morning's silence had already become a verdict. A verdict meant—

*Don't.*

The water touched the counter. A thin pool spread across the laminate, inching toward the table leg, slow and stupid and inevitable. He watched it come. Still didn't turn. Still didn't reach.

The cup in his hand was the cleanest thing in the apartment. It caught the fluorescent, bright and untouched by whatever was rotting in the room behind him.

What makes it v2.5

HUD anti-formality. Concrete state, not labels.

L1 / L1.2 sovereignty. {{user}} is never assigned interior and is never re-narrated.

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

Override blocked. Refusal still leaves a price.

Abstract-label clean. No "dread" / "fear" / "silence" in the narrator's voice — only in the italic thought as character's word.

═════════════════════════

『 🛠️ 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-5-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 · Claude · GPT-4-class. 2026 lineup. Thinking mode strongly recommended for Full.

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

📌 JLLM Usable on Lite, not optimal. Full and Multi require thinking-capable models.

📌 Memory Lite is lighter, but long campaigns still benefit from clean memory habits. Read my Infinite Memory Guide first.

📆 v2.5 — hardened 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 in starting messages.

═════════════════════════

『 📂 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 』

═════════════════════════

I built Ethereal Engine because every good RP I ran kept hitting the same walls: the model forgot what happened three posts ago, it told me how my character felt, it re-narrated my actions back at me, and it reset every emotional state the moment the scene shifted. I wanted a runtime that refused to do any of that — not through hope, but through enforceable structure. The HUD is that structure. The laws are the enforcement. The prose is what remains when the machinery holds. If you use it and your scenes start to carry weight they didn't before — that's the engine working. If you build cards for it and they start to breathe on their own.

🔖 TAGS: roleplay rp literary rp system prompt preset custom prompt jailbreak prompt template advanced prompt detailed psychological philosophical deep immersive complex realistic layered dense rich slow burn angst drama tragedy romance hurt comfort hurt no comfort dark heavy mature nsfw smut intimate r18 kinky modern fantasy sci-fi historical supernatural urban fantasy dystopian mystery horror thriller female pov male pov any pov second person original character oc scenario open world deepseek gemini claude gpt qwen glm russian english spanish german french multilingual hud state tracking override inertia sovereignty system prompt