Workshop · 10 minutes

How the Source Photo Affects an AI Result

A single prompt can give an accurate portrait or a random similar person, and the difference often begins before generation — in light, angle, face scale and number of details that models have to guess.

18 August 20264 source types5 criteriachecklist
One woman is filmed in soft light, backlight, profile and too shallow in the shot
The best source doesn't have to be studio. It's important that the face occupies a prominent part of the frame, the features are not obscured by shadow, and the angle corresponds to the future result.
Brief conclusion

AI is good at transforming the visible and worse at restoring missing things.

A blurry eye can't be made accurate without guessing, a profile doesn' t contain the full shape of the other eye, and a twenty-pixel face doesn’t store small features. The more initial traits available, the less the model replaces a person with an average version.

15–35%Frame width - practical face size
1A soft light source is better than a few hard ones.
4Keep the full facial outline visible
01

Light gives shape, not just makes a picture beautiful.

Soft side or front light shows the volume of your cheeks, nose shape and jaw line. Hard backlight turns these signs into a silhouette; overlit window eats your hair, colored backlight changes your skin tone. The model can compensate for exposure, but often comes up with new texture along with it.

Good.

The face is evenly lit, eyes are readable, the skin is not broken into white.

Allowable.

There is a soft shadow, but the contour and both sides of the face are distinguishable.

Risky.

The face is darker than the background, colored light hides the natural tone.

Bad.

Overlight or black shadows destroyed the details of the eyes and nose.

02

The source angle should be close to the task.

For front-facing business photography, it's easier to use a frontal or light rotation rather than a profile from the bottom. A big shift requires you to build an invisible half of your face. This is no longer editing, but reconstruction, so the risk of loss of similarity increases dramatically.

A wide-angle selfie from close range enlarges the nose and narrows your ears; a high-raised camera changes your forehead and chin.If you want a natural portrait, better shot from arm's length or further away and without the extreme tilt of the phone.

03

Face Size More Important Than General File Resolution

A 4,000-pixel photo can be a weak reference if the person occupies a small corner of the frame. For an avatar, it is more useful to have a clear crop where you see your eyes, nose, mouth, chin contour and hairline. Do not cut off the crown of the head and chin if future composition requires them to be restored.

This is a qualitative selection scale, not a measured percentage of successful generations.

04

The background affects the boundaries of hair, clothing and posture

The contrasting calm background helps to separate the silhouette: branches in the hair, people behind the shoulders and clothes of the same color as the wall create controversial boundaries. When changing the background, the model can trim strands, add a translucent halo or change the arm.

05

Evaluate the source in five questions

  1. Can you see both eyes and their shape?For the frontal result, this is the main source of similarity.
  2. Does the contour of the face read?Hair, ears and chin should not merge with the background.
  3. Is there any lubrication?Especially check your eyelashes, lips and hairline.
  4. Is the future scene suitable?The smaller the turn, the less reconstruction.
  5. Is there a reserve around your head and shoulders?It helps to change the background and framing without finishing the body.
Outcome

Strong source saves attempts better than long prompt

Choose a frame that shows the information you want, not just the most beautiful shot, and then formulate one change and explicitly list what should remain the same.