What Kind of Source Photo Animation Needs
Eighty per cent of the result is determined by the source. No model can extract detail that is not in the frame — it can only invent it, and invention always shows.
Sharpness beats resolution
A 4000-pixel photograph taken with slight camera shake will give a worse result than a crisp 1000-pixel frame. The algorithm relies on edges: the line of the eyelid, the border of the lips, the contour of the chin. Blur those edges and they start swimming once things move.
A practical benchmark is 900 pixels wide or more, provided the face occupies a noticeable part of the frame. If the face is the size of a thumbnail, more resolution will not help.
Light
Soft diffused light from one clear direction works best. Hard backlight, deep crushed shadows and blown highlights create areas with no information — and that is exactly where artefacts appear as the head turns.
Mixed light is a distinct difficulty: a warm bulb on one side, a cold window on the other. The model tries to hold both casts at once and begins to confuse them in motion.
Angle and pose
- A frontal or slightly turned portrait is the best case
- Profile works worse: half the face is unseen and has to be invented
- A lowered head and averted gaze reduce the range of possible movement
- Hair falling across the face is the source of the most visible errors
What to avoid
Screenshots from messengers, over-compressed images with visible blocking, collages, and frames that have already been through heavy retouching. The model reads all of it as genuine detail and faithfully reproduces it in motion.
Go to the site →Read next
Animating Old Photographs: Bringing Movement Back to the Archive 18+ Photo Animation: How It Differs From the Regular Kind Why Faces Swim in Animated Photos, and How to Prevent It Expression, Gaze and Breathing: What Makes an Animated Portrait ConvincingThis material is intended for adults over 18 and is informational in nature.