The model has never met your pet. It only knows the picture.
It does not know how your cat asks for dinner, which corner of the sofa is hers, or that the tip of her tail is white. It gets pixels: what kind of animal, what colour, where the markings sit, the shape of the face, whether the ears stand up or fold down. Anything the photo does not show, it fills in.
How well it fills in depends on how much the photo gives it. So whether the result looks like your pet is mostly decided before you press generate.
How the photo is used
Every template takes one photo.
The page shrinks the image, then sends it to the video service with the scene description. A fixed block in the template prompt asks the model to keep the species and breed, the fur colour and markings, the face shape, the eye colour, the ear shape, the body proportions and the tail. None of that comes from anywhere but the photo.
The model cannot know your cat has a dark patch around one eye unless the picture shows it. If the photo shows it, the render can keep it.
A few hard rules before upload:
- JPG, PNG or WebP
- under 20 MB
- at least 300 pixels on every side
- close to portrait or square; very wide or very tall photos are rejected
Those are gates, not quality. A photo can pass all of them and still tell the model very little.
The four things that matter
The whole head is in frame
Ears especially.
Upright, floppy or folded, ears are the first thing we use to tell one animal from another. Crop half an ear and the model will not leave a gap — it will invent one, usually with the wrong shape, angle and colour. Whiskers are the same: frame too tight and they end at the edge.
Leave a little room above the head.
One animal
With two pets in frame, the model does not ask which one you meant. It picks the larger, sharper, more central animal. The other one may disappear or get folded into the same body.
If you have two, photograph them separately, one video each. A photo of both is a different problem, and not one this photo can solve.
Enough light to see the markings
Markings are the fastest way to recognise your own pet.
Backlight, a dark room lit only by a screen, a black cat on a black sofa — the markings go first. Daylight from a window, falling from the side or slightly in front, is usually enough. No studio light needed; just avoid harsh overhead sun and strong backlight.
Black cats and black dogs are the hardest. Find a window and let the light catch the edge of the fur. That works better than a flash in a dark room.
A normal perspective
Three metres away at eye level beats a wide shot from above.
Shooting down changes proportions: the head shrinks, the body stretches, the legs look short. The model copies those proportions, and you end up with a long neck and a small head.
The phone's standard 1x lens is fine. A 0.5x wide lens distorts the perspective, and the closer you are, the worse it gets.
One more thing people forget: focus
A sharp photo is not the same as sharp eyes.
The model reads the face from the eyes and the nose. If the eyes are soft, the gaze in the render usually is too. Tap your pet's eye on the screen, wait for the focus box to settle, then take the shot.
And avoid images that have been through a chat app. Compression eats fur detail, and dark coats and fine stripes suffer first. Send the original from your camera roll.
What does not matter
A messy background, something on the floor, your hand in the corner of the frame, a collar, warm indoor lamplight — the scene replaces all of it.
No studio, no backdrop, no need for your pet to look at the camera. Looking away, sniffing the floor, a three-quarter profile: all fine, as long as the face and markings are clear.
Common cases
Black cats and dark dogs. Easy to shoot as a black blob. Use side light from a window so the fur catches an edge. Do not fire a flash in a dark room; you will get glowing eyes and nothing else.
White cats and white dogs. Easy to overexpose, and the nose and markings go with it. Avoid bright light in front of snow or a white wall; look for shade or light through a curtain.
Two pets. Separate photos. The model will pick one from a group shot.
Kittens and puppies. Fine. They move fast, so wait for the second they stop and look toward the lens.
Portrait mode. The blurred edges soften ears and whiskers. The normal camera mode is safer.
A ten-second test
Shrink the photo to thumbnail size, or squint at it.
Can you still tell what animal it is, what colour, and roughly where the markings are?
If yes, the model probably can too. If not, take another one.
After upload
One photo, vertical 9:16, with the duration set by your chosen template. Choose the version and quality before generating; the credit cost is shown first.
If the result is off, compare the eyes, the markings and the paws, and see which one drifted. A clearer photo is worth trying, but it is not a guarantee; sometimes the model or the service is having a bad day, and a second run is enough.
Phone photos are fine, by the way. The best results we have seen came from a kitchen, with light coming through the window and the cat turning its head.