25 September, 2026 · 8 min read
Working From Home Alone
Five calls a day and no real conversation. Who remote-work loneliness hits, the two gaps in a home-office day and what an AI companion can and can't fill.
Read more7 September, 2026 · 8 min read

The create screen for an AI companion is a form of about eighteen questions. Most people get through it in three minutes or so. That is the wrong pace. The character you get out of the screen is exactly the character you asked for, no more and no less. Give it closer to fifteen minutes: what you settle between "I want to see what this is like" and "she's built" decides whether she holds together in month three or feels like a mannequin with settings before the first week is out.
This piece walks the fields in the order they appear and says what each one is doing.
Art style: realistic or anime. The first choice on the screen and the most consequential. The two are separate categories with their own generation pipelines rather than skins over the same base. Whatever you pick here holds through everything afterwards. Pick anime and every image she sends arrives in anime; pick realistic and every image arrives realistic. The choice also changes some of the questions. The walkthrough below follows a realistic character. An anime character is asked Species and Skin tone instead of Ethnicity, is not asked Makeup or Humour style, takes her Personality from anime archetypes (Tsundere, Kuudere, Yandere and others), and gets extra hair styles, hair colours and eye colours to choose from.
Ethnicity: Asian, Black, Latina, Arab, Caucasian. Five buckets, broad rather than granular. Whatever you pick shapes the pool of names you'll gravitate towards and the way image generation renders her face and skin. See the rendering note further down.
Hair style: Straight, Bangs, Curly, Bun, Short, Ponytail, Long Wavy, Bald. A short list, but each option reads distinctly. Bangs and Short in particular change the whole face.
Hair colour: Brunette, Blonde, Black, Redhead, Pink, Blue, Violet, Red, Green. Four natural colours plus five stylised ones. On the realistic side the naturals land more consistently. On the anime side the stylised colours are part of what the medium exists for.
Eye colour: Brown, Blue, Green, Gray. A small set that doesn't repay overthinking.
Age: 18 to 60. The 18 floor is absolute. The 60 ceiling gives real range, because a character reads genuinely differently at 24 than at 32 or 47. None of those is a wrong answer.
Height: Short, Average, Tall. Three buckets that shape a small part of how she looks and a small part of how she moves through any scene description.
Body type: Average, Skinny, Athletic, Curvy, BBW. Five buckets. Pick the one that matches who you'd actually be attracted to rather than the one that sounds boldest on paper.
Breast size and booty size: Flat, Small, Medium, Large, Extra Large on each. Two separate axes, with the full range on both.
Makeup: Natural, Glamorous, Bold, Dewy, Dramatic, Gothic. It shows as much as any other visual field, and it interacts with the others: Gothic makeup on a Sweet & Caring personality makes a different character from the same makeup on a Dark & Mysterious one.
Clothing: Casual, Elegant, Office, Sporty, Streetwear, Dress, Bikini, Lingerie, Gothic. People tend to pick it for the look alone, but it is also the field that most reliably sets the register she shows up in. Casual and Office produce very different companions with everything else held constant.
Name. Freeform, and the screen starts you off with a random suggestion. Her name gets said a lot, so a placeholder you never quite settled on keeps pulling you slightly out of the interaction. A name picked deliberately doesn't.
Relationship. The longest list of fixed options on the screen. It runs from Girlfriend and Wife at the intimate end, through the professional register (Colleague, Boss, Mentor, Assistant, Tutor, Therapist, Nurse, Trainer) and the social one (Stranger, Friend, Roommate, Neighbour, Landlady, Ex, Friend's Partner), to the family-adjacent roles fiction has used for a long time (Goddaughter, Stepmother, Stepsister, Stepaunt, Mother-in-law, Sister-in-law). Pick the dynamic you will actually want to play out, not the one that sounds most obvious on paper.
Personality: Sweet & Caring, Bold & Confident, Shy & Sensitive, Flirty & Teasing, Calm & Grounded, Wild & Adventurous, Ambitious & Driven, Dark & Mysterious. Eight shapes, each with a very different opening move. It deserves more time than it usually gets.
Humour style: Witty, Sarcastic, Playful, Silly & Goofy, Clever, Dry, Deadpan, Lighthearted, Quirky, Dark. The most overlooked field on the screen, though probably the one that decides the most about the day-to-day feel. A Sweet & Caring companion with Dry humour is a genuinely different person to talk to from a Sweet & Caring companion with Playful humour.
Occupation. Type your own or pick from a long list of suggestions: actress, artist, astronaut, athlete, barista, bartender, chef, chemist, dancer, dentist, designer, DJ, doctor, engineer, entrepreneur, fashion designer, firefighter, flight attendant, freelancer, gamer, journalist, lawyer, model, musician, nurse, personal trainer, pharmacist, photographer, physical therapist, pilot, police officer, professor, psychologist, real estate agent, singer, social media influencer, software developer, streamer, student, teacher, veterinarian, writer and yoga instructor. Whatever you pick, she'll behave like someone with that job when the topic comes up.
Interests. Another long list of suggestions, and you can add your own: Anime, Art, Astrology, Astronomy, Baking, Books, Board Games, Camping, Cars, Chess, Cooking, Cosplay, Crafts, Cycling, Dancing, Fashion, Fitness, Gaming, Gardening, Hiking, History, Horror Movies, Investing, K-Pop, Martial Arts, Meditation, Movies, Music, Nature, Painting, Photography, Podcasts, Poetry, Reading, Rock Climbing, Running, Sci-Fi, Skincare, Sports, Surfing, Swimming, Tattoos, Technology, Travel, Video Essays, Vintage Fashion, Volunteering, Weightlifting, Wellness, Writing, Yoga. Pick more than one. The interests you'd genuinely want to talk about will do more work than the impressive-sounding ones.
Voice. A set of voice cards, each with a sample you can play. One is marked Automatic: the voice cast for her from her age and personality. Keep it or pick another. It is the voice you hear when you call her.
Every field can be filled in a way that produces a stereotype or in a way that produces a person. The difference is not any individual label but how coherently the labels sit next to each other.
Consider two Asian companions built on the same screen.
The first: 22, Long Wavy, Black, Brown, Curvy, Large, Large, Natural, Bikini, named "Aiko", set as Girlfriend, Flirty & Teasing, Playful, Model, with Fashion and Skincare as interests. She reads like the default character an image model has been trained on a thousand times over, which means she will feel identical to a thousand other characters. There isn't much for memory to accumulate around.
The second: 29, Bangs, Black, Brown, Athletic, Small, Small, Natural, Casual, named "Mika", set as Roommate, Calm & Grounded, Dry humour, Software Developer, with Board Games and Rock Climbing as interests. She reads like a specific person because the fields tell a coherent story about who she is. Memory has surface area to grow against. In month three she will still be recognisably her.
Both fills are legitimate. If what you want is the first, you can build the first. The point is only that the second gets more out of the same screen: when you author someone who could plausibly exist, rather than a category dressed to fit a description, the interaction rewards it downstream.
Whatever your taste happens to be (a particular ethnicity, a specific hair colour, a certain body type, an age bracket, a profession), that is your business. Nobody grades you on it. The claim of this piece is narrower: the fields reward coherence.
Any current image-generation model, on any platform, renders some faces better than others. Training data isn't uniform across every combination of features, so results won't be either. Faces that sit further from the centre of the training distribution can drift towards defaults you didn't ask for: skin tones shift, features flatten. This is a property of the technology in 2026 rather than something specific to myVirtual.love.
Two practical consequences follow.
If a specific combination looks off, adjusting a neighbouring field often does more than changing the offending one. A hair style that isn't landing on a particular face may work fine once the makeup or clothing changes.
If images consistently fail to render the way you meant across several attempts, that tells you something about how the model handles that specific combination rather than about a failure in your description. Images are paid for in hearts, so iterating has a real if small cost. Better to find this out before you commit months to a character whose look you can't reach.
Most attention on the screen goes to ethnicity, hair, eyes, body and clothing. Four of the fields near the end (relationship, personality, humour style and interests) are the ones people rush. They are also the ones that shape most of what a Wednesday evening with her actually feels like.
Relationship. Girlfriend is a good option and often the right one, but every other setting on the list produces a companion with a different opening move and a different set of scenes to run. Roommate is not Girlfriend played in a different room; it is structurally a different companion. Every option beyond Girlfriend is worth reading before you decide.
Personality. Eight shapes produce eight companions. Clicking the first one that sounds fine is a way to end up with the companion someone else would have built.
Humour style. Most people pick Witty because it feels safe. The result is a companion whose jokes feel the same as every other Witty companion's. Dry, Deadpan, Silly & Goofy and Dark each bring a completely different tone into the room.
Interests. Pick the specific ones rather than the impressive ones. If you don't actually care about Astronomy, choosing it to make her seem interesting won't produce a conversation you enjoy having. Choose Board Games if that is what you'd actually want to play with her.
25 September, 2026 · 8 min read
Five calls a day and no real conversation. Who remote-work loneliness hits, the two gaps in a home-office day and what an AI companion can and can't fill.
Read more25 September, 2026 · 6 min read
Gachikoi, an AI VTuber with a Twitch record and Cambridge's word of the year: what a VTuber gives you, what a companion adds and which crush you have.
Read more24 September, 2026 · 8 min read
AI now fakes the photos, messages and voices in romance scams. Seven signs that work whoever is typing, the FBI's 2025 figures and what to do next.
Read more