V4 | Facial Analysis + AI
v4 HTN III
ascend...
7 P/M
1.07 C/L
mirin 6 | 20 |
Mar 30 2026 |
DISCUSSION
LOOKSMAXXING

Facial Analysis + AI

V4


Was bored the other day so figured I would give one of these a shot, uses a local mediapipe model for landmarks + some ratios I pulled off of various forums and studies. Purely bone/math/ratio based, I may implement features later such as eye color, skin tone, hair, skin texture, acne, etc.


Make sure the photo is dead straight with the eyes at a neutral position(the canthal tilt has lot of weight on the score and it will show as asymmetrical if the image is angled.


https://analysis.v4d6.dev



Scoring mechanism

How does it score??

There is a couple values for each ratio/measurement, there is the band(ideal range) as well as the ideal value, and a standard devation. If it is inside the band the lowest score possible is a 75, depending on how close it is to the ideal the score will be higher or lower. Once outside of the range the score goes down more depending on how far it is outside of the band, once it reaches outside of the standard deviation there is an even larger penalty(think of it like the storm from games where it ticks more the further it gets)


Then all of the scores for the measurements are averaged and weighed:

Then for the overall score all of these are averaged, and a .15x worst average score is applied as well as a curve score which is why the 1-10 score is different then just dividing the overall score by 10.


the actual individual scoring still works this way but the penalty section is what actually makes the score


Penalties:

Certain features have more or less weight depending on importance. Also, depending on how many bad features the person has there is another multiplier added. So if they have multiple very bad features(scores below 49 are bad, scores below 25 are very bad), it will add an extra multiplier(negative impact).

(outdated, for me to explain the new system I would need an entire other thread its fucking crazy)



Monetization:

In the future I am planning to monetize this or at least certain parts, extra stuff etc, not gonna leave u guys tho so all of yall will get lifetime if I put it behind a paywall


Contributors:

lookist - Found multiple issues/bugs and also recommended a new ratio("Iris vertical position")

ignazxan - helped with penalty system, ratings accuracy, weights, ratios, etc


feel free to leave any feedback, other measurements I should add, issues, etc

What percentile would 86/100 score be?

Mar 30 2026
lookist
What percentile would 86/100 score be?



...

Top 2-5%, im adding percentiles, MTN HTN etc, as well as making the ratings more accurate since they are currently just a tiny bit off(the individual ratio scores are good but the overall is a bit off)

Mar 30 2026
v4
lookistWhat percentile would 86/100 score be? ...

That's what I got for my post-surgery morph so I'm on that hopium

Mar 30 2026
lookist
v4lookistWhat percentile would 86/100 score be? ...

Updated the site to show the percentile + equivalent rating

Mar 31 2026
v4
lookistv4lookistWhat percentile would 86/100 score be? ...

Lifefuel



I think facial thirds harmony is bugged btw (or the auto placement is)

Mar 31 2026
lookist
v4lookistv4lookistWhat percentile would 86/100 score be? ...

I'll look into this

Mar 31 2026
lookist
v4lookistv4lookistWhat percentile would 86/100 score be? ...

Ah, I see what you mean, its just an animation/drawign issue, the actual math and placement is correct, the drawing just makes it look wrong

Mar 31 2026
v4
lookistv4lookistv4lookistWhat percentile would 86/100 score be? ...

I mean the facial third harmony always seems bugged

Mar 31 2026
lookist
v4lookistv4lookistv4lookistWhat percentile would 86/100 score be? ...

Hmm, alr I fixed the animation/drawing ill look into the actual calculation

Mar 31 2026
lookist
v4lookistv4lookistv4lookistWhat percentile would 86/100 score be? ...

Thanks for feedback, fixed it, should work right now, hairline estimation should also be better now

Mar 31 2026

You should make a feature to edit the points, mirin the site

Mar 31 2026
recessedmandible
You should make a feature to edit the points, mirin the site


...

The points are draggable

Mar 31 2026

You should add a new measurement for dimorphism for a percentage of where the iris sits between the skull top and chin. In women it's ideally below 50% and for men ideally above. Chronic M had a video on it but it's private for some reason.


Because the skull top needs to be approximated in most cases and can deviate due to difficulty of measurement, it could be based on the top of the forehead instead, but new ideal measurements would need to be made.


Mar 31 2026
lookist
You should add a new measurement for dimorphism for a percentage of where the iris sits between the ...

Interesting, I'll def look into adding it, do you happen to have an ideal for men vs female, I'll see if I can find any info on it.

Mar 31 2026
v4
lookistYou should add a new measurement for dimorphism for a percentage of where the iris sits betwe...

It you use the cranium ratio, above 50% is good for men and below is good for women, I don't have an example of the range

Mar 31 2026
lookist
v4lookistYou should add a new measurement for dimorphism for a percentage of where the iris sits bet...

Thanks, I was able to find a bit of info but only for the cranium so I'll do some testing using trichion point instead of cranium

Mar 31 2026
lookist
You should add a new measurement for dimorphism for a percentage of where the iris sits between the ...

Added it, also updated ratings + added a free AI to talk to(make sure your photo is straight as I heavily penalized bad scores and if its angled you will get ~30 on canthal tilt which will fuck your score)

Apr 1 2026
v4
lookistYou should add a new measurement for dimorphism for a percentage of where the iris sits betwe...

I think the range for the measurement may be too large, still annoying I can't access the video I'm mentioning which goes into more detail.

Apr 1 2026
lookist
v4lookistYou should add a new measurement for dimorphism for a percentage of where the iris sits bet...

I’ll take a look

Apr 1 2026