AI models for greetings, from photo cards to talking videos

Picture models: faces, lettering or the whole family

For a photo card with your own people, start with Auto, which is GPT Image 2.5 Flare from OpenAI. It keeps likeness best of the picture models we run and finishes a card in about 25 seconds. Its sibling GPT Image 2 is the one for cards where the words matter: the watercolour card above reads "Happy 70th, Mum" with every letter in place. It takes about 80 seconds, so it is not the model for quick redos.

When the whole family has to fit, switch to Google's Nano Banana Pro. It keeps up to five faces recognisable and drew all 36 scene pictures in our catalogue. Seedream 5.0 Pro from ByteDance goes for rich colour, like the sunlit garden dinner above. FLUX.2 Pro aims at photo realism, Nano Banana 2 is the fast one, and Qwen Image 3 writes clean text but takes only three reference photos, so it suits smaller groups.

Video models: seconds, sound and what a second costs

A scene video is made in two steps, the card and then the animation, so it takes three to six minutes. MiniMax H3, our Auto for video, animates all 20 video scenes, always adds sound and costs less per second than Kling 3.0 or Veo 3.1. "Make a wish, Mum" above is H3: candles out in one breath, clapping, a hug.

Kling 3.0 Pro from Kuaishou gives smooth motion and the longest clips here, up to 15 seconds at 1080p, sound on or off. In "The whole family raises a glass", glasses clink and Grandma wipes a happy tear. Veo 3.1 from Google has the cinema look and makes voices in the same generation: in "Mum blows out the candles" the family sings the end of Happy Birthday. It stops at 8 seconds and costs more per second; Veo 3.1 Fast and Lite are cheaper. For a calm loop, Wan 3.0 Prime from Alibaba holds the frame steady.

Songs, voices and talking cards run on fixed engines

Not every format has a Model picker. Songs are always made with Google Lyria 3.5: the agent writes the lyrics from what you told it, or you paste your own, you approve them, pick a genre and a mood, and only then is the song made. The words can be in English, Russian, Hebrew, Arabic and more, and all 40 songs in our catalogue were made with it.

Voice greetings are read by ElevenLabs Eleven v3 in one of 8 catalogue voices. It is the only voice model here that covers Hebrew, so a Hebrew greeting always goes to it. On a paid plan you can record a short phrase once, and from then on MiniMax Speech reads your greetings in your own voice.

The talking card has two settings instead of a model list. Better is Kling AI Avatar v2, which moves the face and head with the words; Best is OmniHuman 1.5, which moves the whole person and lets you point at who speaks in a group photo. Either way you get up to 15 seconds of speech, paid from the same balance as everything else. Text greetings cost nothing, and a new account starts with 1,000 free credits.