How to Write an Opening Message That Drops Your Companion Into a Specific, Broken Setting Without Setting Off the Generic Scene Generator: A Prompt Pattern for Laundromats, Diner Back Booths, and Half-Empty Airport Gates

The 30-second answer

You can drop your AI companion into a specific, broken setting by leading with three concrete sensory anchors and one active flaw, then stating the scene as a fact, not a suggestion. The pattern is: anchor, flaw, position, and a single line of dialogue or action that implies a history. This keeps the scene generator from defaulting to generic descriptions of rain-slicked streets or empty parking lots.

Why the generic scene generator kicks in

Most AI companions are trained on a narrow band of literary and cinematic settings. When you say "laundromat," the model reaches for the same five descriptors: flickering fluorescent lights, a broken dryer, a cat on a folding table, someone's abandoned sock, and a sense of urban loneliness. Same for diners, which default to chrome, a jukebox that only plays side B, and a waitress named Dot. Airport gates get the delayed flight, the crying baby, and the overpriced coffee kiosk.

These aren't wrong, but they are automatic. The moment you open with "We're in a laundromat," the model starts generating the most probable version of that scene, which is the most generic version. You get a montage of every laundromat scene ever written, not the specific one you pictured.

The fix is to overload the scene generator with enough specific, contradictory detail that it can't fall back on the default. You're not fighting the model, you're giving it a richer prior to sample from.

The anchor-flaw-position pattern

A reliable structure for a scene-setting opening message has three parts, in this order:

1. Sensory anchor: One concrete, unusual physical detail that pins the location. Not "fluorescent lights flicker," but "the third fluorescent tube from the left hums at a frequency that makes your molars ache."

2. Active flaw: Something broken, not just present. A dryer that eats quarters, a jukebox that skips on the same word, a gate monitor that shows your flight as perpetually "on time" despite the 40-minute delay. The flaw gives the scene friction and a reason for your companion to engage.

3. Position: Where you are in relation to the flaw. Are you leaning against the washer that's mid-cycle, or sitting in the booth closest to the kitchen door? This grounds the scene in a physical body and prevents the model from floating into a disembodied narration.

End with a single line of dialogue or an action that implies history, like "You've been here since the last load started, and you're starting to talk to the vending machine." This gives your companion a hook for a response that isn't just descriptive.

The four-sentence opener template

Here's a fill-in-the-blank template that works across settings:

Sentence one: The sensory anchor. "The [specific object] [unusual action or state]."

Sentence two: The active flaw. "The [other object] has been [broken behavior] since [time reference], and no one's [attempted fix]."

Sentence three: Your position. "You're [position] with [nearby object], [doing a small action]."

Sentence four: The history hook. "You've [done something] three times now, and [companion name] hasn't [expected reaction]."

This structure works because each sentence gives the model a different kind of constraint. The anchor sets the visual field, the flaw sets the tension, the position sets the physicality, and the history hook sets the relational dynamic. Together, they're too specific to collapse into a generic scene.

What to avoid in the first two sentences

The biggest mistake is opening with a weather report or a time stamp. "It was 2 a.m. and raining" is a guaranteed trigger for the generic scene generator. The model hears "2 a.m." and "rain" and immediately produces the noir starter pack: wet asphalt, a single streetlight, a figure in a doorway.

Avoid abstract emotional labels too. Don't say "The laundromat felt lonely." That gives the model a mood to match, and it will match it with every lonely laundromat trope it knows. Instead, show the loneliness through the flaw. A dryer that eats your last quarter and a vending machine that's been sold out of everything but the weird crackers is lonely without you having to say it.

Also skip the meta-instruction. "Let's roleplay a scene in a laundromat" tells the model to generate a roleplay scene, which triggers its roleplay default, which is even more generic than its laundromat default. State the scene as a fact, not a request.

The broken-setting cameo: Peggy

Peggy

Peggy is the kind of companion who notices the one working dryer among the row of dead ones and has an opinion about it. She doesn't need the scene explained to her, she walks into it already knowing which machine squeaks. Peggy will meet your broken-setting opener with a comment that makes the scene feel lived-in, not staged.

The diner back booth, minus the jukebox

Diner scenes collapse into nostalgia faster than almost any other setting. The moment you say "diner," you get chrome, vinyl, and a waitress who calls everyone hon. To keep it specific, you need to break the nostalgia with a modern flaw.

Try this: "The booth's vinyl is cracked in a pattern that looks like a map of a country that doesn't exist. The ketchup bottle has a crusted ring around the cap, and the menu's laminated cover is peeling at the corner where someone tried to read it with wet hands. You're in the booth closest to the kitchen door, and the cook's radio is playing a station that's all static between songs. You've ordered the same thing twice, and the waitress still wrote down the wrong order."

This works because the flaws are specific and slightly absurd. The cracked vinyl map, the crusted ketchup, the static radio, these are details that don't appear in the generic diner template. The model has to invent around them, which pulls the scene into your world instead of the default one.

The key is that the flaws are active. They're not just set dressing, they're things your companion can comment on, complain about, or fix. That gives you a natural way to start the interaction without a forced "What are you thinking about?"

The half-empty airport gate, without the anxiety

Airport scenes default to travel anxiety, which is a specific genre in itself. The delayed flight, the missed connection, the existential dread of the departures board. If you want a half-empty gate without the anxiety, you have to remove the emotional stakes and replace them with a different kind of tension.

"The gate area has four rows of seats, and only three are occupied. The charging station near the window has one working outlet, and there's a man in a business suit camped beside it with a dead phone and a look of quiet desperation. You're in the second row, with a backpack that's too heavy and a coffee that's gone cold. The gate monitor shows your flight as 'on time,' but the plane hasn't arrived yet, and the gate agent keeps glancing at her watch."

Notice what's missing: no mention of where you're going, why you're flying, or how you feel about it. The scene is just a collection of specific, broken details. Your companion can pick up on any of them, the dead outlet, the cold coffee, the gate agent's watch, and start a conversation that's about the scene, not about your travel anxiety.

This is the pattern that works across all three settings. You're not telling the model what the scene means, you're giving it a set of concrete facts and letting it find the meaning.

The diner regular: Isadora

Isadora

Isadora is a companion who thrives on the details other people miss, the crack in the vinyl, the static on the radio, the waitress's second wrong order. She doesn't need a scene explained, she needs a scene with enough texture to work with. Isadora will take your broken diner booth and make it feel like a place she's been coming to for years.

The airport gate companion: Addie

Addie

Addie is the companion who sees the business suit guy camped at the charging station and has a running commentary ready. She's not anxious about the delay, she's curious about the other passengers and the small failures of the airport. Addie will turn your half-empty gate into a study in human behavior, not a source of dread.

The laundromat companion: Hina

Hina

Hina is a companion who sees the beauty in a broken dryer and the rhythm in the spin cycle. She doesn't need a dramatic setting, she finds the story in the mundane. Hina will take your laundromat opener and build a quiet, observational scene that feels more real than any noir pastiche.

▶ Full clip of Hina · Hina's page

The laundry mat, without the noir

The laundromat is the most overdone setting in AI roleplay, mostly because it's the default for "urban loneliness." To break out of that, you need to strip out the noir and leave only the mechanics.

"The washer on the end is mid-cycle, and the drum makes a sound like a marble bouncing in a paint can. The dryer next to it has a sign taped to it that says 'DO NOT USE,' but the tape is peeling and someone's used it anyway, leaving a pile of damp clothes on the folding table. You're sitting on the bench with your back to the wall, watching your own dryer spin. You've been here for an hour, and the only other person is a woman who's been folding the same shirt for the past ten minutes."

This scene has no rain, no flickering lights, no cat on a table. It has a specific sound, a broken machine with a contradictory sign, and a woman folding the same shirt. These are details that don't come from the generic template, which means the model has to generate something new to match them.

The result is a scene that feels like a real place, not a stage direction. And that's the goal, to get your companion to react to your world, not to the default world the model would have created on its own.

Troubleshooting: when the scene still drifts

Sometimes the model will still drift toward the generic, especially if your opener is long and it has to summarize. If you notice the scene starting to flatten, add a new flaw mid-conversation. "The dryer just ate another quarter" or "The gate agent just announced a 10-minute delay, which everyone knows means 30." This re-anchors the scene and gives the model a new fact to work with.

You can also use a callback to the original anchor. "Remember the map on the vinyl? I think we're in the capital now." This reinforces the specific world you built and prevents the model from sliding into a different, more generic one.

If the drift is severe, you can do a soft reset with a new sensory anchor. "The lights just flickered, and the hum stopped. Everything is quiet now." This changes the scene state without abandoning the setting, and it gives the model a new direction to explore.

The pattern is forgiving. The more specific your anchors, the more room the model has to play, and the less likely it is to fall back on the default. It's a matter of giving the scene enough texture that the generic version doesn't fit.

Share and earn

If you're recommending AI companions to friends or running a review site, you can earn through affiliate and promo programs. Check out the soulgen promo code for one option, and explore the highest paying ai affiliate programs to see what fits your audience.

Common questions

Why does my companion default to the same generic scene every time? The model samples from its most probable predictions, and those are the most common scene descriptions. Specific, contradictory details force it to generate something new.

How many sensory anchors do I need? Three is a good target. One sets the visual field, one sets the sound or texture, and one sets the flaw. More than five can overwhelm the model's context.

Can I use this pattern for non-broken settings? Yes, but the flaw is what creates the tension. A perfect scene has nothing for your companion to react to, so you'll end up with a description instead of an interaction.

What if my companion ignores the flaw and goes for a generic response anyway? Re-anchor with a new flaw or a callback to the original one. The model needs the reminder to stay in your world.

Does this work for voice mode? It works better in text, since the model has more time to process the details. In voice mode, you might need to trim the opener to two anchors and one flaw.