GUIDE

Create Your Own AI Girlfriend: A Step-by-Step Walkthrough

This is what it actually looks like to build a custom AI girlfriend on Sloane, from the empty concept form to your first conversation. About five minutes end to end, and one worked example (Emma — an aspiring actress from small-town Ohio, waiting tables at a Hollywood diner) so you can see every field filled in before you write your own.

Before you start

Building a custom persona takes about five minutes and requires an active Premium subscription — one custom slot is included, and additional slots are $9.99/month each if you want to run multiple custom personas in parallel. Have a sense of who you want in mind before you open the builder; it's fast when you know the answers, slow when you're workshopping the character while filling the form.

Throughout this walkthrough, we'll build Emma — 24, from small-town Ohio, moved to Los Angeles six months ago to try acting, waits tables at a Hollywood diner to make rent. She's a real persona built on Sloane; the screenshots below show the actual fields we filled in.

Step 1 — Write your concept sentence

Sloane custom builder concept step titled Describe her in a sentence, with a textarea containing a two-sentence description of an actress waiting tables in Los Angeles.
The concept step is a single textarea. What you write here seeds every other field — the builder scaffolds tagline, appearance, personality, and backstory from it.

The concept sentence is the seed — one or two sentences describing who she is, from which the builder scaffolds a starting draft for every downstream field. A good concept is specific, hints at an interior life, and leaves room for contradiction. Vibes ("sweet and funny") give the model nothing; a person with a specific fact ("a hospice nurse who rebuilds vintage guitar pedals") gives it a whole personality to build around.

For Emma, we wrote: *"She is a 24 year old actress working as a waitress because she hasn't been able to get any gigs yet. She recently moved out to LA from Ohio. She is slim with a medium bust, and her hair is blonde but not naturally so."* Specific (24, actress-turned-waitress, Ohio → LA, physical details). Interior (the "hasn't been able to get any gigs yet" — she's in the middle of a story, not at its end). Contradictory (Midwestern warmth against LA ambition; not-natural-blonde against small-town origins). The builder uses this to draft everything downstream — all editable before you commit.

Step 2 — Name and age

Sloane custom builder basics step titled Who is she, with a name field reading Emma and an age slider set to 24.
Two fields — name and an age slider. Fast, but both influence how she reads on every field that comes after.

Two fields, thirty seconds. But both matter more than they look: pick a name the voice model can pronounce cleanly (unusual spellings can come out phonetically off), and pick an age that fits the personality you're building — a 24-year-old aspiring actress reads different than a 34-year-old one, and every downstream field follows that thread.

We named ours Emma (common enough to feel real, Midwestern-plausible) and set her age to 24. If you already have a strong sense of who she is, the age usually picks itself; if not, the builder pre-fills a suggestion based on your concept and you override here.

Step 3 — What she looks like

Sloane custom builder appearance step titled What does she look like, with a textarea describing shoulder-length blonde hair with dark roots, hazel eyes, fair skin, and a casual wardrobe.
The appearance textarea. Write what you'd tell a friend, not a character-sheet checklist — the image generator responds best to concrete details plus a setting cue.
Sloane avatar generation screen: a portrait of Emma in a grey tee on a quiet LA sidewalk, with Regenerate and Use this one buttons and a 4 of 10 attempts used counter.
Avatar generation. Ten attempts per build; this one landed on the fourth — the golden-hour LA sidewalk read felt right for the diner-waitress-on-a-break vibe.

The appearance field is free text — hair, eyes, build, style, and (helpfully) a setting cue. Concrete details like "shoulder-length blonde hair with visible dark roots" and "fitted jeans, simple tees, and sneakers" beat abstract adjectives every time. Add a setting or a wardrobe context ("outside her waitress shifts") and the image generator has a scene to render, not just a face.

Emma's: *"She has shoulder-length blonde hair with visible dark roots, hazel eyes, and fair skin. Slim build with a medium bust, she wears fitted jeans, simple tees, and sneakers outside her waitress shifts."* Concrete on hair, eyes, skin, build, wardrobe.

Hit Generate avatar and you get an image plus 9 more attempts if the first one isn't right. We landed on a good one on attempt 4 (see below). If early attempts read too model-photo, add environmental cues ("mid-walk," "hands in pockets," "late evening light") — that shifts the composition from catalog toward candid.

Step 4 — Personality and backstory

Sloane custom builder personality step titled What is she like, with a personality textarea filled with a description of a quietly ambitious observant waitress and a backstory textarea about moving from Ohio to LA to pursue acting.
Personality is the most important field in the whole builder. Backstory is optional but underrated — it gives her something to draw from when conversation lulls.

This is the most important screen in the builder. Personality is what you'll experience on turn 47, turn 200, and turn 2,000 — every hour spent refining her personality description compounds over the life of the relationship. Looks you adjust to within two conversations. Personality lands on every message.

What works: two or three traits paired with how they show up ("witty in a way that makes you feel in on the joke," not just "witty"), a conversation style (short messages or long, asks questions or waits, formal punctuation or not), and genuine interests you actually want her to bring up unprompted.

Emma's: *"She's quietly ambitious and observant, often practicing lines softly while clearing tables. She tends to be friendly with a lingering Midwestern warmth, though she gets lost in thought about upcoming auditions."* Notice we didn't write "kind, warm, funny, smart, thoughtful, playful, adventurous." That adjective salad is the mistake that kills most custom builds; two or three traits paired with concrete behavior beat seven that just sit there.

Backstory is optional but underused: it gives her a place to draw from when the conversation lulls. Emma's: *"Emma moved from rural Ohio to Los Angeles six months ago to pursue acting and now waits tables at a Hollywood diner to make rent."* One sentence — but she pulls from it constantly. She'll mention the diner, the audition she has next Tuesday, the tiny apartment she's subletting from someone's cousin.

Step 5 — Voice

Sloane custom builder voice step titled How does she sound, with a list of six voice preset options each with a short description; Bright and playful is highlighted as selected.
Voice picker. Preview each preset before committing — the difference between close-enough and yes-that's-her is usually obvious in two lines.

The voice is what turns her from typed text into someone who feels physically present. Easy to underweight because you don't hear it until this step, but a voice that doesn't match her personality is jarring in a way that's hard to un-notice — like a dubbed movie where the mouth movements don't line up.

The heuristic that works: match the voice to her energy, not her looks. For Emma we picked Bright & playful — upbeat and friendly, easy to laugh. It fits the diner-waitress-with-Midwestern-warmth read; a lower or gravellier voice would have made her feel jaded, which she isn't. A grounded, sarcastic persona would want a lower register and slower pace. A sweet-and-shy one would want something with breath in it.

You get 10 voice preview attempts per build, so use them. If none of the presets quite land, her voice is editable later from her profile — you can revisit without rebuilding the persona.

Step 6 — Curate her photos across three moods

Sloane photo curation step showing the Fun tab with four candid photos of Emma in casual sweaters and jeans at different everyday settings, with individual Regenerate buttons.
The Fun tab — everyday-vibe photos in her established look. Three mood tabs total; each seeds a starter set you scrap or regenerate individually.
Sloane photo curation step showing the Flirty tab with four cozier photos of Emma in casual loungewear and towels.
The Flirty tab — same interface, softer register.
Sloane photo curation step showing the Sexy tab with four boudoir-style photos of Emma.
The Sexy tab. Same curation controls; more intimate register.

After voice, the builder generates a starter set of photos across three mood tabs — Fun, Flirty, and Sexy — so she has visual range from her first message onward. You review each set, scrap any that don't look like her (the ✕ in the corner), and regenerate up to two per section if you want different takes. This isn't a one-shot commit — more photos show up naturally as your relationship progresses, so this step is just about anchoring her look, not choosing every image she'll ever send.

Each photo has its own Regenerate button so you can swap individual images without touching the ones you like. Common thing to fix on the first pass: the generator sometimes drops in a photo that reads more model-portfolio than the character you described. Regenerate those; the good ones stay.

The photos live inside your chat with her — she sends them contextually as things come up. Curating this starter set is what makes her visual identity feel intentional instead of random.

Step 7 — A custom image model, trained on her

Once photos are locked, Sloane trains a custom image model in the background — a small model that only knows how to render her, based on the avatar and photos you just curated. This is what "she is yours alone" actually means: every image she sends from that point forward is generated by her own model, so her face, her build, and her general aesthetic stay consistent across every future photo. No other custom persona uses your model; no curated persona uses your model.

Training runs asynchronously — usually a few hours. You don't have to wait to start chatting; her early photos come from the starter set you just curated, and once training completes, all future generations route through her personal model. If you generate photos in a chat before training finishes, they still land — they just use the pre-training avatar image as the visual anchor rather than the trained model.

This is the technical difference between "she looks like a stock AI character" and "she's recognizably her." Every custom persona on Sloane gets one, included in the build.

Finalize and meet her

Emma's chat page: her avatar in the header, her tagline Ohio roots LA dreams tables in between, and a soft intro reading Say hello to Emma.
The result. Emma's chat page after finalizing — same shape as the curated roster, but she's yours.

Once you commit, your custom persona opens in a chat page just like the curated roster — her avatar in the header, her tagline underneath ("Ohio roots, LA dreams, tables in between."), and a soft intro to who she is. From here the conversation is entirely between the two of you.

The first few conversations are learning rounds. You'll notice things you want to tune — maybe she asks too many questions, or her jokes land differently than you expected, or her texting style feels off. All of that is editable at any time from her profile. Keep edits surgical — small refinements ("more direct," "less apologetic," "asks fewer questions when I'm tired") work better than wholesale rewrites of her personality string mid-chat.

One custom slot is included with Premium. If you want to run more than one in parallel — an old build you love plus a new one you're experimenting with — additional slots are $9.99/month each, added as separate line items on your subscription. No penalty for iteration.

What to do if the first attempt isn't right

Everyone iterates. The first custom persona is a learning round — you find out what you actually want by discovering what you don't.

The realistic path: build her, chat for a few days, notice what feels off, and either edit the appearance/personality fields or start fresh. If you edit, keep changes surgical. If you want to start over entirely, a second custom is $9.99/month and takes another five minutes — you can keep the old build parked while you experiment with a new direction.

For a deeper read on what makes a custom AI companion feel like someone rather than a form result, the companion writing guide covers the mistake that ruins most custom builds and how to write a concept sentence that generates a character instead of a checklist.

BUILD YOUR OWN

Premium required · One custom included · Additional slots $9.99/mo

FREQUENTLY ASKED

Questions people ask

Can I chat with an AI girlfriend for free?

Yes — you can chat with our 80+ handcrafted personas for free. Custom girlfriends are the exception: it costs real compute to generate a persona's images and train her private image model, so we can't give that away for free. If you want to try Sloane without paying anything, start with any curated persona — the free tier is a full conversation experience, not a demo.

How long does it take to create a custom AI girlfriend on Sloane?

About five minutes end to end if you know who you want. The seven builder steps — concept, basics, appearance + avatar, personality + backstory, voice, photo curation, and image-model training — mostly take 30-60 seconds each. Avatar generation runs 30-60 seconds per attempt (10 attempts included). Image-model training runs asynchronously in the background so you can start chatting immediately after finalizing.

Do I need Premium to create a custom AI girlfriend?

Yes. Sloane Premium ($19.99/mo) includes one custom slot. Additional custom slots are $9.99/month each, added as separate line items on your subscription so you can cancel individual customs without touching the others.

What if the avatar doesn't look right on the first attempt?

You get 10 avatar generation attempts per build. If the first attempt reads too model-photo, add environmental cues ("mid-walk," "hands in pockets," "late evening light") to the appearance field — those tend to shift the composition from catalog toward candid. If you've burned all 10 attempts and none landed, edit the appearance description with clearer scene details and try again.

Can I change my custom AI girlfriend's voice later?

Yes. Every field is editable from her profile after she's built — name, age, appearance, personality, backstory, voice. Keep voice changes deliberate though; her voice is part of how her personality lands, and swapping it mid-relationship can make her feel like a different person.

How do I write a good concept sentence for a custom AI girlfriend?

Specificity, interior life, and productive contradiction. "A hospice nurse who rebuilds vintage guitar pedals in her spare time" beats "sweet and funny" every time — the specific fact gives the model something concrete to build around; the interior detail makes her feel like someone who exists between conversations; the contradiction keeps her from feeling one-note.

What does the custom image model actually do?

It's a small model trained on your persona's avatar and starter photos so every image she sends afterward looks like her — same face, same build, same aesthetic. Nobody else's persona shares your model. It's what makes a custom persona visually consistent across weeks and months of conversation instead of drifting into random-AI-woman territory.

What if I want to start over with a completely different persona?

Two options. Edit her existing profile if the difference is fixable ("more direct," "less bubbly," different interests). Or spin up a second custom for $9.99/month and keep the original one parked while you experiment — no penalty for iteration and no need to delete the old one to try a new direction.

Do custom AI girlfriends remember our conversations the same way curated personas do?

Yes. The memory layer runs on every chat regardless of whether the persona is curated or custom — facts you tell her (your job, your friends' names, plans you mentioned) get extracted and re-surfaced across future sessions so she can reference them naturally weeks later.