Make a photo talk: the birthday card Grandma says herself

Standard or Best: how much of the picture moves

Both engines get the same two things from us: the card, and the voice, already recorded and trimmed to the seconds you paid for. The difference is how much of the picture moves. Kling AI Avatar v2, from Kuaishou, works on the face and head: the mouth follows the words, the head moves with them, and the rest of the photo stays as it was, from the clothes to the room behind. For a portrait of one person that is usually exactly right, and it costs less per second. The grandmother on this page was made this way.

OmniHuman 1.5, from ByteDance, animates the whole person: head, face and shoulders move with the rhythm of the voice, so a half-length shot or a raised glass gives it more to do than the mouth. The grandfather on this page, speaking Russian in the Ari voice, was made with it from one portrait. It also solves the group-photo problem: switch the quality to Best, tap the face of the one who should talk - Grandpa thanking everyone, say - and he speaks while everyone else stays in the picture, listening. Without a tap the model chooses the speaker, which is fine for a portrait and a gamble for a crowd, so keep the table small: a tiny face at the far end is hard to tap. Because it costs more per second, Kling AI Avatar v2 is the everyday choice.

Mouth in view, face to the camera

The face that talks is the face on the card, so likeness starts there. GPT Image 2.5 Flare, our Auto choice for pictures, has the best likeness, and cards with up to four people keep faces most reliably. For the talking part, a few things help: the face turned towards the camera, the mouth in view rather than behind a cup or a hand, even light, and a good share of the frame for the face. For OmniHuman 1.5, keep the shoulders in view as well. One person in a portrait is the easy case; a tiny face in a wide landscape is the hard one.

If your only photo is a group shot from across the garden, make a new card first, closer in, and let that one talk. People saves you a step: when Grandma is saved there with a photo, the agent already has it and does not ask again.

What fits into fifteen seconds of speech

Fifteen seconds is a toast, not a letter: the name, one real detail, one wish. The grandmother's line on this page, "Happy birthday! I wish I could hug you today - this is the next best thing.", runs 7 seconds, so the limit holds about two lines like it. The voice pauses at full stops, and a sentence that runs for three lines comes out as one long breath. "Happy birthday to my amazing grandson, wishing you all the best on your special day, may all your dreams come true" says less than "Happy birthday, Noah. Eight already, and you still owe me a rematch at checkers. I'm bringing the cake on Sunday."

Use the name the person is really called: Nana, Dedushka, Saba, Teta. Keep Hebrew and Arabic lines a little shorter, since 15 seconds holds fewer letters in them than in English. Skip stage directions like [laughs]: the app strips out anything in square brackets before the voice reads the line. The best words usually come from the person in the photo, so call Grandma, ask what she would say and type it her way. Then read the line aloud once; if it trips you up, it will sound stiff in their mouth too.

Choose the voice for the listener, not for yourself: a slow, warm one for a grandparent, a bright one for a six-year-old. Each catalogue voice has a sample, so choose by ear rather than by name. And one rule that is not technical: use photos of people who would be glad to see themselves talk, and words they would actually say. Put a phrase in Grandpa's mouth that he would never use, and the family will notice before the candles are out.

For whoever cannot be in the room

Most talking cards stand in for someone who is missing. Grandma is two flights away on the day, so her portrait says happy birthday from a phone propped against the cake. If she has her own account on a paid plan, she records her voice once, and then it really is her speaking.

Distance is not the only reason. Filming yourself saying happy birthday takes a dozen attempts, and the good one has your thumb in it; here you pick a photo you already like and type two sentences. Time zones make birthday calls awkward, so a card with your face, and on a paid plan your own voice, can wait for them: schedule it for their morning, not yours.

Language counts too. The grandchildren answer in English while Babushka still thinks in Russian. Write her line in Russian, the way she says it, and her photo speaks it in Russian, because the voice reads whatever language the line is written in.

Better as a video, a voice or a song

A talking card is one person, a few sentences, looking at the camera. It is not a party. If you want the whole table raising glasses, with clinking and laughter, that is a video. Kling 3.0 turned our long-garden-table card into a 5-second clip of exactly that, sound included. Veo 3.1 makes voices and singing in the same generation - on its page, a family sings the end of Happy Birthday and Mum blows out the candles. MiniMax H3 animates every video scene in our catalogue and is the best value of the three.

If the voice matters and the face does not, send a voice greeting: the same eight voices, or yours, with no picture to animate, and it costs less. If you want the words sung, Lyria 3.5 sings lyrics you have approved, in English, Russian, Hebrew, Arabic and more.