How to Take a Great Phone Photo for an AI Portrait

About a quarter of the AI portrait previews we generate could have looked twenty percent better with a slightly different source photo. The customer rarely realises it, because the preview still looks good, but the room between good and great is almost always closed at the camera, not at the model. Here is the brief I would tape to the fridge if I had to hand it to every customer before they upload.

Light is doing eighty percent of the work

A phone photo taken next to a window in soft daylight will outperform a phone photo taken under overhead room lights every single time. The reason is that side light gives a face shape, depth, and a believable highlight in the eyes. Overhead light flattens everything and leaves dark sockets where the eyes should be. If you can rotate the subject so that one cheek is closer to a window than the other, you have already won.

Angle and distance

Hold the phone at roughly the same height as the subject’s eyes. Photos taken from above tend to flatten foreheads and shrink jaws. Photos taken from below add chins. For pets, get on the floor. A photo taken at the dog’s eye level beats every photo taken from a standing human’s perspective. Distance matters too. Step back a metre or two and zoom with your feet rather than the digital zoom slider. Phone digital zoom degrades the file in ways that show up later in the portrait.

Expression and gaze

Posed grins are weaker than they look on the camera roll. Slight smiles, gentle expressions, and gazes directed slightly off-camera read more honest in a portrait. For pets, try squeaking a toy or saying a familiar name just before the shutter, the ear flick captures personality. The single hardest thing to fix in post is a closed-mouth grin that feels braced. Aim for the second after laughter rather than the laughter itself.

Background

The AI will redraw the background in almost every style, so a busy background is fine, but it does affect the lighting on the subject. A bright background (white wall, kitchen window) lifts the whole image. A dark background (wood-panelled hallway, dim corner) drops it. Pick one based on the mood you want. Watercolor likes bright, oil likes moody, anime does not care.

What to avoid

• •Direct overhead sun outdoors, especially around midday. Adds harsh shadows under eyes and chin.
• •Flash. Flattens skin, glares pet eyes, never improves a portrait.
• •Heavy filters or beauty modes baked into the camera. They strip detail the AI needs to render features accurately.
• •Group photos cropped to one person. The crop usually loses pixels the AI wanted.
• •Old or compressed photos pulled from a chat app. Always use the highest-resolution original from the camera roll.

For pets specifically

Pets are harder to direct, so the rules tighten. Get to eye level. Side window light. Treat or toy held just above the lens to catch the gaze. Capture the full face, including both ears. Cropped ears or muzzles confuse breed-specific renderings. For long-haired breeds, brush before the shoot. The AI cannot un-mat a coat, and matted fur reads as flat colour in the portrait.

The five-second pre-upload check

• •Is there a catchlight (a small bright dot) in the eyes? If yes, light is good.
• •Can you see both eyes clearly? If no, retake.
• •Is the face larger than a thumb on the screen? If no, get closer.
• •Are there harsh shadows splitting the face in half? If yes, move out of direct sun.
• •Does the expression look like the person you know? If yes, upload.

When the photo cannot be retaken

Memorial portraits, old family snaps, photos of pets who have passed, these are the cases where you cannot just shoot again. In those situations, choose the style that forgives the photo. Watercolor and Ghibli are the most forgiving. Oil and anime are the least. If the original is small, dark, or blurry, ask for watercolor and let the medium do the softening work that the photo cannot.