The look-suggestion engine, end to end

How a look gets picked for your day

From the first onboarding question to the outfit on your avatar. Every module, what it knows about you, and how it decides, explained without code. Tap any module to look inside.

Live in the app Built, switched off Planned

Overview

Five modules learn about you. Ten modules build the look.

The outside modules collect and tidy what the app knows. The inside modules turn it into one outfit for today. Tap any box for the details.

Outside modules: what the app knows

Inside modules: how the look is built

Overnight

At 03:00 local time the engine picks the look for the new day, dresses your avatar and sends a morning notification.

Opening Home

If no overnight look is waiting, Home asks the same engine right away.

Asking the stylist

"What should I wear tonight?" uses the very same engine, plus your words.

1Outside module · Onboarding Live

Eleven short steps that teach the app your life

Each answer becomes a fact the engine can use, or a rule it must never break. Pick a step to see what it asks and what it turns that answer into.

Body shape, build and skin tone are used only to draw your avatar. The suggestion engine never reads them.

2Outside module · Wardrobe intake Live

From a phone photo to a garment the engine understands

Photoyou snap it

Checkis it usable?

Cut outbackground off

Readtype, colour, details

Confirmonly if unsure

Clean shotredraw if needed off

Step 2: the photo check

An AI looks at the photo, and the app measures the pixels itself. The app makes the final call, so a "looks fine" from the AI cannot hide a dark photo.

Steps 3 to 5: what it reads

Who wins when two readers disagree:

YouVision AIClothing tablesShop tags

Anything below 60% certainty (type, colour, warmth or formality) is shown to you to confirm.

3Outside module · Stylist chat memory Live

It remembers what you said, and only what you said

The stylist can offer to save a preference. The app then checks that your exact words really appear in your message. If they don't, nothing is saved.

Where a memory goes

Avoid colour / itemNever in any suggestionLive

Prefer, facts, feedbackStylist's knowledge only

Plans and eventsReshape the day's timeline

Nine places the stylist can search

ProfileGarmentsLooks and ratingsMemoriesChatEventsPostsCommentsTry-ons

Each question pulls the 8 best matches. You can switch off chat, posts and comments as sources.

4Outside module · Understanding requests Live

Any language, any slang, turned into a clear plan

You can write a note the way you'd text a friend. A fast AI turns it into simple instructions, and each one has to quote your own words. Pick a note:

The dressiness scale, 0 to 7

The scale is about clothes, not events: every garment is rated on it. The AI decides where your occasion sits, and the look must land within one step of that (two on a uniform day).

Calendar titles stay on your phone

Event names like "Team dinner" are classified on the phone in 8 languages and never sent to the server. The onboarding step that connects the calendar is still a placeholder.

5Outside module · Geo and weather Live

The weather where you'll actually be, hour by hour

Where you are

Your shared location, or your region's centre if you didn't share it.

A 5 km weather cell

Nearby people share one forecast. It is kept for 1 hour.

Live forecast

Open-Meteo worldwide, the national weather service in the US. The app waits at most 2 seconds.

Backup plan

If the forecast fails, the last good one counts for 3 hours. More than 7 days ahead, the app uses typical weather for about 700 regions.

Rooms too

Indoor temperature follows the day outside, between the local heating level and 24.5°C.

A 12°C day in Yerevan

Each time you are outside, the engine plans for the coldest hour of that stretch, not the day's average.

Rain counts from 20% chance Wind counts from 3 m/s, fully at 8 Heat rules start above 30°C

Yerevan · day

Your Home avatar stands in front of your city: a real picture for each time of day, and rain wins over the hour. Only Yerevan's 5 backdrops are live so far.

6Inside modules · Day segments and warmth Live

Your day, cut into stretches, each with its own warmth need

Tap a stretch. The engine knows where you are, for how long, and how hard you're moving. The bar shows the warmth that stretch needs and what the outfit gives.

Rule: a short dash outside never needs a coat

Under 5 minutes doesn't count. From 5 to 20 minutes it counts step by step, and from 20 minutes it counts fully. Car rides count partly, up to an hour.

Minutes outside4 min

Temperature outside-5°C

Simplified to show the rule. The real engine uses international comfort standards for cold, mild and hot weather. "Warmth points" is the standard insulation unit: 1 point is roughly an ordinary business suit.

7Inside modules · Occasion and hard filters Live

First decide how dressy. Then take out what can never be worn.

Where the dressiness target comes from

The engine walks down this list and stops at the first answer it finds. Then it allows one step either way, or two on a uniform day.

The hard filters

These are not scores. A garment that fails one is simply removed before any outfit is built.

8Inside module · Building candidates Live

Thousands of combinations, narrowed to the three worth showing

The funnel

If you ask for something you don't own ("must be pink"), you still get a look, and the gap is named: "nothing you own is pink".

Dressing rules: tap layers on and off

Layers go on in a fixed order, from skin outwards. The engine never builds an outfit that breaks these rules.

Shoes and accessories come last, and never change which clothes win

Rain bootsfavoured at 50%+ rain, if you own them

Sandalsranked lower under 15°C or at 40%+ rain

Scarfat 10°C or colder, never from 20°C

Hat and glovesat 5°C or colder

Sunglassesin the summer months or above 30°C

9Inside module · Scoring

How the engine weighs warmth, dressiness, colour and the rest

Two designs exist. The live one sets hard limits first, then blends the rest into one score. The newer one, built but switched off, settles things like a ladder: rule by rule, in strict order.

Step 1: hard limits

Live

Yes-or-no checks remove whole outfits before anything is scored. None of them can be outvoted.

Step 2: one blended score

Live

A dry 12°C office day, with no note and no loved colours set. Rain and requests don't apply, so they drop out and the rest are rescaled. Move the weights and watch the ranking.

In a blend, factors can trade off: push colour up and warmth down far enough, and the light blazer outfit wins even though it's colder.

10Inside module · Colour, part of scoring Live

Colours are judged by how far apart they sit on the colour wheel

Close colours are calm, opposite colours are bold, and the middle distance clashes. Neutrals like grey, navy and camel go with almost anything. Move the second colour.

First colour

Second colour

Colour match score

0.70

Apart on the wheel

160°

What hides counts less

The outfit's colour score weighs each pair by how much of it people can see.

2%

shirt under a coat

10%

shirt under a jacket

40%

shirt under a knit

11Inside modules · Variety, pick, final check, explanation Live

Three looks that are good, and different from each other

Pick the second look: a close copy or a real alternative?

Each alternate is chosen by 70% "how good is it" minus 30% "how much does it look like what's already picked".

How different two looks are

40%different top

25%different coat

15%different bottom

20%different colours

Who makes the final pick

No note: the highest score is the main look.
With a note: a fast AI reads up to 5 numbered looks, described in words without photos, and picks one. It has 8 seconds.
Too-cold looks are dropped while at least 3 others remain. Turned-down looks can be kept out too; the server is ready, no screen sends them yet.

Final check, then the explanation

A last check confirms you own every piece and the outfit is complete. If something's wrong, it's an error, never a quiet patch. Shortfalls are said out loud.

Warm enough for 22 min outside at 9°C. The outer layer comes off indoors.

Fits the business-casual occasion. The colours sit well together.

Take an umbrella.

12Inside module · Try-on on your avatar Live

The chosen look is drawn onto you, with live progress

Selfie

Your avatar

Dressed

What you see while it works

WeatherPickingMatchingQueuedDrawingCheckingOn you

Home starts dressing once the look has stayed the same for 1.2 seconds. If drawing fails, you see "Try again". It never retries quietly in a loop.

How the picture is made

Pieces go on in dressing order: dress or bottoms, top, middle layer, outerwear, shoes, accessories.
A fast image AI draws them onto your neutral grey-base avatar.
A second AI looks at the result: is the grey base showing, or is a garment missing?
If something is wrong, it redraws once and keeps the better picture.
Each look is drawn once and remembered, so opening it again is instant.

3 try-ons a day free30 with premiumA failed try-on gives the credit back

13Self-learning

It grows its own knowledge, on a strict budget

Nobody writes lists of activities by hand. The first time anyone mentions "padel" or "a friend's henna night", the app works out what that means for clothes, saves it, and improves it later.

How a new activity is understood

Runs; feeds the ladder engine (off)

A small daily AI budget for learning

Over budget, unknown activities stay "unknown" until tomorrow, so the bill never surprises anyone.

When an entry gets improved

A person says the kit was wrong, or a reviewer marks it wrong.
The app wasn't sure (under 70%) when it wrote it.
A sport, ceremony or place of worship has no clothing rules yet.
It's been used 25 times and nobody has reviewed it.

At most 2 automatic rewrites per entry, 6 hours apart. Reviewed entries are never overwritten. One entry can hold up to 24 local versions, for example "wedding guest in China: no white".

What your reactions change today

Try it yourself

Twelve everyday situations the engine is tested on

Pick a situation and watch the decision play out step by step. Each one is a real automatic test that runs on every change.

Pick a situation

The steps will appear here.

Safety and limits

Fast, cheap AI where it helps. Plain rules everywhere else.

The suggestion core itself is ordinary rules and maths, so it's predictable and free to run. AI is called only to read photos and free text, to chat, to pick between close looks when you wrote a note, and to draw.

Fast models only

Only quick, low-cost AI models are allowed. The expensive class is refused everywhere.

Strict time limits

Forecast 2 s. Look pick 8 s. The stylist gets 6 steps, 25 s each, 50 s per answer, and stops after 2 failures in a row.

Daily budgets

3 free try-ons and 3 free studio photos a day. Learning has its own small daily budget, and every AI call is counted.

Kill switches

The stylist, try-on, the overnight run and live weather can each be switched off in seconds.

No invented facts

The stylist can only show looks and sources it actually saw in that turn. Memories must quote your words.

Respectful selfie use

From a selfie, only skin colour, hair colour and hair style are read, never age, gender, ethnicity or body. If you allow it, the photo also guides how your avatar’s face is drawn.

Checked answers

A final check rejects any look with a piece you don't own. Errors are reported, never hidden.

Graceful fallbacks

No forecast? Use the last good one, then typical weather. No AI pick? Use the score order, when that is allowed.

What's next

Built and waiting, or planned

Everything above marked "Live" works in the app today. These pieces are either finished and waiting for the switch, or still on the plan.

Get early access Back to LookFreedom