You upload an old black and white photo of your great-grandfather, and less than a minute later he gives you a faint smile. It looks like magic, or like a simple filter from a phone app.
In reality, those roughly 45 seconds hide an enormous amount of work that would not have been possible at all just a few years ago. So how does bringing a photo to life actually work, and why does it take so much computation?
Why it is not just a "stretched picture"
Animation used to rely on something called morphing. The computer took the pixels of the original photo and mechanically stretched them like rubber. The result looked flat and unnatural, and people ended up looking like wax figures.
Modern neural networks take a completely different approach. They do not see an image as a flat grid of coloured dots. Instead, they try to understand it in space:
- Where exactly is the depth of the scene, and where does the light on the face come from?
- What does the face look like from a slight three-quarter angle, even though the photo was taken straight on?
- What is hidden behind a shirt collar, and how does the fabric of a coat crease when the person takes a gentle breath?
In other words, the model has to imagine the whole scene in three dimensions while working from a single flat photograph.
Step 1: Cleaning up and "reading" the photo
Old photos have their charm, but for algorithms they are a tough nut to crack. They often arrive as phone snapshots of a family album: slightly crooked, with a yellowed border, glare from a lamp or a scratch across someone's face.
Before anything starts to move, automated systems have to prepare the photo carefully:
- 1Separate the photo from its surroundings: The algorithm has to recognise exactly where the actual paper print ends and the table or scanner begins, so that a frame on the wall or the pattern of a tablecloth does not start moving.
- 2Anchor the facial features: The system maps dozens of reference points on each face. It has to remember precisely that the man in the picture has a thin moustache, fine lines around his eyes or a distinctive nose, so the generator does not turn him into a stranger as he moves.
The better the photo is captured, the easier this step becomes. If you are choosing what to animate, our guide on which photo works best for animation will help.
Step 2: Powerful chips calculate dozens of complete frames
Once the scene is ready, the most demanding part begins: generating the motion itself.
This does not happen on an ordinary home computer or on your phone. The request travels to a specialised data centre running extremely powerful computing chips. A single one of these chips costs about as much as a new car, and its only job is to perform a huge number of calculations every second.
The model does not generate "a video as a whole". It calculates dozens of separate, fully detailed frames one after another. For every single frame it has to work out again:
- how the shadows under the chin have shifted,
- how the reflections in the eyes are preserved,
- how millimetres of movement in the shoulders, head and eyelids flow smoothly into each other.
If the model got even one frame slightly wrong, the viewer's eye would instantly notice it as a jerk or an unnatural distortion.
Step 3: The biggest challenge – human warmth instead of a robot
The hardest part of the whole process is not the movement itself, but making it natural and subtle.
Basic generators tend to overdo things. People start wildly turning their heads, blink at regular intervals like machines, or everyone in a group photo moves at exactly the same second.
For the result to feel warm and dignified, the instructions have to guide the model towards relaxed micro-movements:
- a gentle breath,
- a faint smile that starts at the corners of the eyes,
- natural asymmetry, with each person in the photo reacting in their own moment.
Why a one-time payment?
Many apps in the app stores run on subscriptions. They tempt you with a "free" trial, but you have to enter your card and activate a weekly or monthly fee (often $7 to $20 a month) that keeps renewing automatically, even long after you have stopped using the app.
Our approach is built on fairness:
- No signup and no app to install: You do not need to create an account or leave your details anywhere.
- No hidden subscription: You will never be charged a recurring payment.
- A one-time price of $2.99: You pay only for the computing time you need at the moment you want to bring a particular family memory to life.
The result is a few seconds of pure emotion. The moment your great-grandfather, or your mother as a young woman, looks at you again just the way you remember.