How to take the right photo for a neural network | MioLook

Over 12 years of working as a colorist and image consultant, I've pored over thousands of client photos. And do you know where 90% of color type misdiagnosis starts? With yellow bathroom tiles. Just last week, a woman approached me, absolutely certain she was a typical "Warm Autumn." She'd uploaded her selfies to various apps, and the algorithms consistently returned terracotta and mustard palettes. But when we met in person in daylight, I saw a classic cool "Winter" with porcelain skin. Why did this happen? Because her smartphone camera dutifully captured the yellow light of incandescent bulbs, and the algorithm analyzed precisely those pixels.

If you want to know, How to take the right photo for a neural network , you'll have to forget about posing for 10 minutes and think about the physics of light. We explain in more detail why algorithms are mathematically more accurate than the human eye in our complete guide: MioLook's neural network will be able to determine a person's color type. In this article, we'll explore the technical side: how smartphone cameras lie to us about our proportions and tones, and how to provide an AI stylist with a flawless source image.

Pixels Instead of Emotions: Why Neural Networks See Us Differently

Our brain is a brilliant color correction tool. If you walk into a room with blue walls or a red lampshade, your eyes will adapt in a split second (the so-called chromatic white balance kicks in), and you'll still perceive a white sheet of paper as white. A camera sensor can't do this. It simply records the raw RGB codes (the proportions of red, green, and blue) of the light waves that bounce off your face and hit the lens.

"Color deceives us constantly," wrote color theory classic Josef Albers in his seminal book, "The Interaction of Color" (1963). And in the age of artificial intelligence, this rule works both ways.

There is a term in coloristics metamerism — an optical phenomenon in which the same color appears completely different under different light sources. Your skin literally changes its physical hue to the camera sensor depending on the light source above you.

The main rule when working with any AI stylist is the GIGO principle (Garbage In, Garbage Out). The neural network doesn't infer your actual skin tone. It takes specific pixels from the uploaded image. Give it a yellow photo and you'll get a wardrobe for yellow skin.

Lighting is the boss: how not to fool the algorithm

According to the PANTONE Color Institute, the ideal color temperature for evaluating shades is considered to be D65, which roughly corresponds to daylight on an overcast day at midday. In digital terms, this represents a color temperature in the range of 5000–5500 Kelvin. Any deviation from this standard will cause the algorithm to make errors.

Ideal light: diffused daylight

For the AI to accurately calculate your contrast level and undertone, you need soft, diffused daylight. The ideal scenario is to stand near a north-facing window on an overcast day. North exposure provides the most even, neutral light without the harshness of direct yellow sunlight.

Stand facing a window, about 0.5–1 meter away. The light should evenly illuminate your face, leaving no harsh shadows under your nose or eyes. Avoid direct sunlight—it acts like a spotlight: it whitens prominent parts of your face, casts shadows into pitch black, and completely destroys the nuances of your natural coloring.

Bathroom and ring lights: sponsors of the wrong color types

Now, here's what you absolutely shouldn't do. Incandescent light bulbs in apartments typically have a color temperature of 2700–3000K. This is a decidedly yellow-orange light. Taking a photo in the living room in the evening will artificially add a warm, peachy undertone to your skin. Office fluorescent light (around 4000K with a strong green spectrum), on the other hand, will make you look sallow and gray.

But the biggest myth in the industry is ring lights. It would seem that every beauty blogger uses them! However, for AI analysis, a ring light is the worst enemy. It produces a harsh frontal light that blurs the natural contours of the face, flattens it, and leaves a bright white highlight in the pupils. Because of this, the algorithm can't correctly calculate your contrast level (the difference between skin, eye, and hair color).

Proportion distortion: how to pose correctly for a neural network

When it comes to body analysis, optics come into play. All modern smartphones are equipped with wide-angle lenses with a focal length of 24–28 mm. The problem with them? They inevitably stretch objects close to the lens and narrow those further away.

If you take a selfie in a full-length mirror, holding the phone at chest level and tilting it slightly, you'll end up with a disproportionately large head, broad shoulders, an elongated torso, and short, tapering legs. AI will analyze this photo and honestly suggest clothes for an inverted triangle body type, even though you might be a classic hourglass figure in real life.

How to restore natural proportions:

• The 2-meter rule and 2x zoom. Mount your phone on a tripod (a standard tabletop or floor tripod for €15–20 will do the job perfectly). Step 2–2.5 meters away from the camera. To avoid looking like a speck of sand in the frame, turn on your camera's 2x optical zoom (50mm equivalent). This focal length is what photographers call a portrait focal length—it conveys body proportions as the human eye sees them.
• Lens height. The phone should be positioned exactly at your belly button or waist level. Never at eye level!
• Vertical. Mount your smartphone at a 90-degree angle to the floor. Any tilt of the device (even 5 degrees) will instantly distort the length of your legs in the photo.

Fair Limit: This zoom rule doesn't work if you're shooting in dim light at night. In low light, your smartphone will switch from optical to digital zoom, causing pixels to crumble, making the image blurry, and the neural network won't be able to detect your contours at all.

A Face Without Filters: Preparing for a Color Type Analysis

To accurately determine your coloring, the AI needs to see your natural skin texture. Any makeup—even the lightest BB cream, a drop of under-eye concealer, or tinted lip balm—changes pixels. Be sure to wash your face before shooting.

Hair should be pulled back into a sleek ponytail or bun, especially if it's been dyed. Your color type is determined by the genetic makeup of your skin and iris, not by the shade applied by your hairdresser. Artificial hair color will confuse the algorithm.

But the most insidious thing is this color reflex One of my clients once sent me a photo of herself without makeup, in perfect daylight, but wearing a neon pink T-shirt. The analysis showed a cool pink undertone to her skin. Why? The bright fabric acted as a reflector: the light hit the T-shirt and cast a pink spot right on her jawline and chin. To avoid this, be sure to wear a neutral gray, white, or cream T-shirt. As a last resort, just drape a white towel over your shoulders.

The Body in the Frame: How to Capture Your Figure for the Perfect Outfit

According to a study by technology agency WGSN (2024), virtual fitting room algorithms make 40% of style selection errors solely due to poor posture and baggy clothing in user photos. If you're standing in an oversized hoodie, the AI doesn't have the X-ray vision to figure out exactly where your waist is.

How to prepare for a figure photo:

1. Cloth. Wear form-fitting black or dark gray leggings and a contrasting, form-fitting top (or camisole). Your uniform should be easy to read.
2. Pose "A". Stand up straight with your heels together and your toes slightly apart. Let your arms hang down by your sides, but relax them. be sure to move it away from your hips by a couple of centimeters There should be some space between your arms and your waist. If you press your arms to your sides, the AI will merge them with your torso, creating a non-existent extra volume at your waist.
3. No posing. Don't stand on your "working side," don't stick out your hips like Instagram models, and don't suck in your stomach to the point of rib pain. The program needs your real anatomical frame. If you suck in your stomach, the AI will recommend pants that will make it hard to breathe after lunch.
4. Background. Find the blankest, most monochromatic wall in your home (ideally, one without colorful rugs, cabinets, or bright baseboards). The more contrast you have against the background, the faster and more accurately the neural network will extract your silhouette.

Checklist: 5 Steps to the Perfect MioLook Source

Let's recap. To get premium consultation-level results from your AI stylist, follow these 5 simple steps:

• Step 1. Wait until daylight (preferably lightly cloudy) and go to a window. Avoid dark rooms and ring lights.
• Step 2. Remove your makeup, pull your hair back with a hair tie, and put on some basic, form-fitting clothing (or drape a neutral white towel over your shoulders if you're just filming your face).
• Step 3. Place the phone exactly at stomach level (for a figure) or at eye level (for a portrait), strictly vertically.
• Step 4. Move 2 meters away from your smartphone and turn on 2x zoom to avoid the fisheye effect.
• Step 5. The most important thing is to go to the camera settings and force-disable all beauty filters, skin smoothing, and face enhancements.

A good starting photo is 99% of the success of working with artificial intelligence in the style field. By spending just 10 minutes creating a technically sound, distortion-free shot, you'll forever free yourself from the agony of choice and end up with a wardrobe that complements your appearance, not clashes with it.