Picking a face for your AI companion used to mean scrolling through a gallery of pre-made characters and settling for the one that was "close enough." On Svila, when you create your own companion, her portrait is generated for her specifically, not pulled from a shared pool. This is the team behind Svila.io writing about how portrait generation actually works today, including the parts that are still rough.
Portrait generation sits at the intersection of two things people care about with AI companions: how the companion looks, and how much say you have over that look. The honest summary is that you have real influence through what you write when you create her, and less control after that than you might expect.
This post walks through how the feature works, where its limits are, and how to get a portrait that matches the character you have in your head.
On this page
- Why the portrait matters
- How we think about portrait generation
- Your Description
- Filling the Gaps
- One Face, Then One Portrait Per Mood
- Consistency Across Moods
- Portrait Boundaries
- Tied to the Companion
- 7. What Isn't There Yet — No Presets, No Re-rolls
- How we approached this
- How to get the most out of your companion's portrait
- Final thoughts
- FAQ
- A note from the team
01—Why the portrait matters
Why the portrait matters
A companion's appearance is one of the first things that makes her feel specific rather than generic. Text conversation builds personality over time, but a face gives you something immediate to anchor to.
It also matters because a portrait that doesn't match the character you described is a common frustration with AI companion apps. You write "red hair and glasses" and get something else entirely. Much of the work behind Svila's portraits went into making the description you write actually count.
02—How we think about portrait generation
How we think about portrait generation
A few things mattered most when we designed this feature:
- Your words first — anything you say about her appearance should win over anything we'd pick at random
- Variety — companions shouldn't all end up with the same idealised face
- One person across moods — her portraits for different moods should read as the same person, as far as the technology allows
- Clear limits — portraits are of adults, clothed, and never sexual
- Tidy deletion — when a companion goes, her portraits go too
The rest of this post walks through the pieces that come out of those priorities.
Your Description
Where the Portrait Starts
There is no separate image prompt box. When you create a companion, the form asks for a name, a description, gender, tone, a personality mood, language and, on Premium, story worlds. The description is optional, and it does two jobs: the appearance details you include shape her portrait, and the personality details shape how she talks.
Svila reads the description for the details that define a face: heritage, face shape, hair length, texture, colour and styling, eye shape and colour, build, and things like glasses, tattoos, piercings or a beard. If you write an appearance detail the system doesn't specifically recognise, such as an unusual hair colour, it tries to use your own wording rather than replacing it with a random pick.
- What it does
- Turns the appearance details in your companion's description into a portrait.
- Best for
- Anyone who already has a specific look in mind.
- How it works
- The description is read for facial and styling details, which go into the image request ahead of anything filled in automatically.
- Why we built it
- A portrait that ignores what you wrote is worse than no portrait at all.
- Bottom line
- The more specific your appearance details, the closer the result.
Filling the Gaps
Why Companions Don't All Share One Face
Most people describe personality, not bone structure. Left to itself, an image model tends to fall back on one polished, symmetrical, interchangeable face whenever a description doesn't say otherwise.
So for every detail you didn't specify, Svila picks one from a deliberately broad range, one feature at a time, and saves those choices with the companion. Leave the description blank and the whole face is chosen this way. Mention green eyes and only the eyes are fixed; everything else is still varied.
- What it does
- Fills in any facial details you didn't describe, choosing from a wide range.
- Best for
- People who care more about who she is than exactly how she looks.
- How it works
- Each unspecified feature is chosen on its own and stored, so the same companion always resolves to the same set of features.
- Why we built it
- Without it, most companions converged on the same generic face.
- Bottom line
- Say what matters to you; the rest is filled in, and it won't be the same face every time.
One Face, Then One Portrait Per Mood
Portraits are generated automatically, in the background, right after you create a companion. You land straight in the chat, and her picture appears once it's ready.
Generation starts with one reference portrait: a plain, relaxed, daylight picture that serves as her neutral look. Then one more portrait is made for each other mood her tone can reach. Svila has three mood states (neutral, playful and flirty), and the mood during a chat is set automatically from her own replies, capped by her tone. That cap decides how many portraits she gets: a Friendly companion has one, Romantic and Sweet companions have two, and companions with a premium tone have three. As the mood of the conversation shifts, the portrait shifts with it.
- What it does
- Creates a reference portrait plus one portrait for each mood her tone can reach.
- Best for
- Anyone who wants the picture to follow the feel of the conversation.
- How it works
- The mood portraits change expression, pose and lighting; the description of her face stays the same.
- Why we built it
- A single static picture can't follow a conversation that moves from easy to playful.
- Bottom line
- One to three portraits, depending on the tone you chose.
Consistency Across Moods
Honest About Drift
Keeping one face stable across several generated images is one of the harder problems in personalised image generation, and we'd rather be straight about where we are.
Each mood portrait is generated from the same saved face details and the same random seed as her reference portrait. That keeps a lot constant, but it does not guarantee an identical face. Her playful or flirty portrait can drift from the neutral one: a slightly different jawline, hair falling differently, a face that reads as a close relative rather than the same person. Stronger face-matching is something we want, and we're not going to claim it before it's there.
- What it does
- Anchors every mood portrait to the same face details and seed.
- Best for
- Setting expectations before you meet her playful or flirty side.
- How it works
- Shared face details and a shared seed, not a guaranteed face match.
- Why we built it
- It's the most consistency the current setup can deliver without overclaiming.
- Bottom line
- Same face details every time, but not a guaranteed identical face. Drift can happen.
Portrait Boundaries
Adults, Clothed, Never Sexual
Svila.io is for people aged 18 and over, and mature themes are welcome. It does not generate sexual content, in conversation or in the portraits it draws.
In practice, every portrait is an upper-body picture of an adult. Every image request includes instructions that block nudity and underwear, and the flirty portrait gets its charge from a look and the lighting, not from clothing coming off. The age of the face is always set within an adult range, no matter what the description says. If a description suggests a character under 18, portrait generation is blocked outright.
- What it does
- Keeps every portrait adult and clothed.
- Best for
- Anyone who wants to know where the line is before creating a companion.
- How it works
- Nudity is blocked in every image request, face age is fixed to an adult range, and descriptions that suggest a minor are refused.
- Why we built it
- The limits should hold for every portrait, not depend on how a description happens to be worded.
- Bottom line
- Attraction can show in a look. Nothing more.
Tied to the Companion
Storage and Deletion
Her portraits belong to the companion, not to a loose folder of images. Companions you create aren't shown to other users, and neither are their portraits. If you delete a companion, her stored portraits are deleted along with her, and so is her Memory Journal.
Companions can't be edited after creation, only deleted, so there's no case where you rename her or rewrite her description and end up with a portrait that no longer fits. If you want a different look, the way to get one is to create a new companion.
- What it does
- Keeps each companion's portraits attached to her, and removes them when she's deleted.
- Best for
- Anyone who wants a simple answer to "what happens to the pictures?"
- How it works
- Portraits are stored under the companion they belong to and cleaned up when she's deleted.
- Why we built it
- Images of a companion shouldn't outlive the companion.
- Bottom line
- Delete her, and her portraits go too.
09—7. What Isn't There Yet — No Presets, No Re-rolls
7. What Isn't There Yet — No Presets, No Re-rolls
Two things people often expect aren't part of Svila today. There are no style presets: portraits are photographic in style, with no anime or illustrated option to pick. And there's no regenerate button: you can't re-roll a companion's portrait or adjust individual features after she's been created. If a portrait doesn't finish loading, "Try again" checks for it again; it doesn't start a new portrait.
That makes the description you write at creation the main lever you have, which is why the next section focuses on it. Crafted companions such as Aria come with their portraits already made.
What it does: Sets expectations about what you can and can't change. Best for: Anyone deciding how much effort to put into the description. How it works: The portrait is generated once from the details saved at creation. Why it's this way: Her face details are saved with her, so nothing re-rolls her appearance by accident. Bottom line: Get the description right up front. That's where your control is.
10—How we approached this
How we approached this
We built portrait generation around one principle: what you write should shape what you get, and anything you leave open shouldn't collapse into the same generic face. Most of the effort went into reading descriptions properly and filling gaps with real variety, not into a menu of options.
We've also tried to be clear about the rough edges. Faces can drift between mood portraits, there are no presets, and you can't re-roll a portrait. We'd rather you know that before you create a companion than find out afterwards. If you want to see how it works, Try Svila free and create a companion of your own.
11—How to get the most out of your companion's portrait
How to get the most out of your companion's portrait
- Put appearance details in the description: hair length and colour, eye colour, build, heritage, glasses, tattoos. Specific beats vague.
- Keep appearance and personality both in there: appearance shapes the portrait, personality shapes how she talks, and the same field does both.
- Leave things open on purpose: if you don't mind about a feature, don't mention it, and it will be varied rather than defaulted.
- Expect some drift between moods: her playful and flirty portraits may not match the neutral one exactly.
- If the look isn't right: companions can't be edited, so create a new one with a more specific description and delete the old one if you like.
12—Final thoughts
Final thoughts
Portrait generation looks simple from the outside: write a description, get a picture. What decides whether it feels right is underneath: whether your words are respected, whether companions look different from each other, and whether the limits hold every time. We think Svila gets the first two mostly right today and holds the third firmly, and we're open about the drift that's still there.
FAQ
Can I choose a style, like anime or realistic?
No. There are no style presets. Portraits are generated in a photographic style from the appearance details in your description.
Can I regenerate my companion's portrait if I don't like it?
Not today. There's no re-roll or edit option after creation. If you want a different look, create a new companion with a more specific description.
Will my companion look exactly the same in every mood portrait?
Not guaranteed. Each mood portrait uses the same face details and seed, but faces can drift between images, sometimes noticeably.
How many portraits does a companion get?
One per mood her tone can reach: a Friendly companion has one, Romantic and Sweet have two, and the premium tones have three.
Will Svila generate sexual or nude portraits?
No. Svila.io is for adults, and mature themes are welcome. It does not generate sexual content, in conversation or in the portraits it draws.
What happens to the portraits if I delete the companion?
They're deleted along with her, and so is her Memory Journal.
14—A note from the team
A note from the team
This post is written by the team behind Svila.io. The features and choices we describe are ones we designed and shipped — so our perspective is first-party, not neutral. We try to be honest about the trade-offs, but you should always try things yourself and form your own view.
Last updated September 2026.
Ready to try Svila.io?
Create your first AI persona for free — no credit card required.
Get Started Free