How to Photograph Your Cat for an AI Video

The photo is the entire input

When a video model turns two photos into a clip, it does not invent a cat. It copies the one you gave it: the fur pattern, the markings, the shape of the face, the color of the eyes, the proportions. Whatever is in the photo is what shows up in the video, and whatever is missing gets guessed at.

That is the whole argument for spending two minutes on the photo instead of ten seconds. A sharp, evenly lit photo produces a cat that looks like your cat. A soft, backlit, half-hidden one produces something cat-shaped that your friends will politely agree resembles yours.

Light: window light, and nothing else

You do not need a studio. You need a window and the right time of day.

• Put the cat facing the window, not in front of it. Light should fall on the face. A cat sitting with a bright window behind it becomes a silhouette, and a silhouette has no fur pattern to copy.
• Turn the flash off. On-camera flash flattens the face and produces two bright discs where the eyes should be. Both are things the model then has to interpret.
• Avoid mixed light. A warm lamp on one side and daylight on the other gives the fur two different colors, and the model has to pick one.
• Mid-morning and late afternoon are easiest. Midday sun through a window is harsh and puts stripes of shadow across the cat; overcast light is soft and forgiving.

Angle: at eye level, facing the camera

• Get down to the cat's height. Photos taken from standing height look down at the cat and stretch the body. Kneel or put the phone on the floor.
• Both eyes visible. A three-quarter view is fine; a cat twisted away with one eye hidden is not.
• Ears in frame. Ears are part of the silhouette - cropping them off changes what the model thinks the head looks like.
• Whole body in frame for the clapping photo. Head, chest, front paws and hind legs. The render stands the cat up and claps its front paws, so the front half needs to be clearly visible.

Sharpness: this is the part that is hard

The model cannot recover detail that the photo does not contain. A slightly blurry photo gives a slightly wrong cat, and a photo taken from three meters away gives a cat with no texture to copy.

• Tap to focus on the eyes before you shoot. On most phones this also locks the exposure.
• Get closer, then crop later. Two or three feet away fills the frame with the cat, which is far more useful than a wide shot of the whole room.
• Shoot a burst. Cats move. Twelve frames of a burst usually contain one where the head is still and the eyes are open.
• Never screenshot a photo. Screenshots lose resolution and pick up the interface. Find the original file.
• Skip heavy filters. Beauty filters, heavy sharpening and skin-smoothing effects all change what the face actually looks like, and the model copies what it sees.

The five mistakes that get a render refused

Most failed renders are not the model misbehaving - they are the input tripping a check. The automatic photo check reads the first photo as a person and the second as an animal, and it needs to be able to see both clearly.

1. Two animals in one frame. The most common one by far. One cat, on its own.
2. A face that is covered. Sunglasses, a hat pulled low, a hand over the face, a mask - anything the check has to guess behind.
3. A heavily filtered or edited photo. If the face has been smoothed into something that no longer reads as a photograph of a person, it gets refused.
4. The wrong photos in the wrong slots. The first slot is the person, the second is the cat. Swapping them is a contradiction, not a variation.
5. A photo of a screen or a printed photo. Photographing a photo adds glare, distortion and a color cast. Use the original file.

Failures are refunded automatically, so a refusal costs you time rather than money - but it is faster to fix the photo than to press generate again.

The person photo, in one paragraph

Everything above applies, minus the tail. One person, facing the camera, evenly lit, nothing covering the face, nothing bright directly behind the head, and no second face in the background. Head and shoulders or half body both work. Group photos are tempting and always a bad idea, because the model has to work out which of the faces it is meant to be copying.

If your cat will not cooperate

This is the real obstacle, and it is solvable:

• Shoot when the cat is already settled. A cat lying down in a patch of sunlight is easier to photograph than a cat being chased around a room.
• Use the burst, not the timing. Take twenty frames and choose afterwards.
• Let someone else hold the toy. A second person with a toy behind you gets the ears forward and the eyes up, which is the expression you want.
• Take the photo from further away and crop. If the cat will not let you get close, use zoom or step back and crop later - just do not go so far that the fur loses all texture.
• Come back later. A photo of a bored cat is better than no photo, but a photo of an interested one is worth waiting an hour for.

Try it with the photos you have

The generator takes the two photos and does the rest: no prompt, no settings, one render costs 100 credits. The full walkthrough - including which slot each photo goes in and how to add the sound - is in how to make a cat clapping video. And if you want to know why the format exists at all, start with what cat clapping is.