Before your grandma's birthday, you suddenly realize: a regular store-bought card is a template. You want warm words, a drawing that will touch her heart. But you don't know how to draw, and a designer will charge several thousand rubles for a card with a picture and will still make it a template. Sound familiar?
In fact, a beautiful personalized card can be put together in an evening without any artistic skills. A neural network will generate an image based on your description, an AI twin will help with the text and design, and you'll add something of your own. This article is a step-by-step breakdown with examples, mistakes, and realistic timelines.
Why regular cards don't work
A purchased card loses not because of the printing, but because of the lack of connection to the person. Grandma sees a cat in flowers and thinks: "Well, a cat is just a cat." But if the picture shows her favorite teapot with blue cornflowers and the caption "Your mint tea is the best," that's already personal.
The second problem is the text. The greeting "I wish you happiness, health, and success" is not a greeting, it's a formality. What matters to Grandma is that you remembered how she taught you to bake pancakes or how you planted potatoes together. A neural network can help put together warm words, but without your details they'll remain generic.
Step 1. Come up with the foundation: what will be in the picture
The neural network won't guess what exactly is important to your grandma. Describe the scene as you would tell a friend. Bad prompt: "beautiful card for grandma." Good: "watercolor drawing: an elderly woman in glasses knitting a scarf, next to her on the table a mug with a daisy on the side, snow outside the window, warm lamp light."
The more specific the details, the fewer reworks. Remove from the description everything that has nothing to do with grandma: the neural network may add extra objects if you haven't limited the scene.
Step 2. Generate the picture — and don't get upset the first time
Modern image generators (for example, via Replicate with the Flux model) produce a result in 10–30 seconds. But the first time almost always comes out wrong: either the face is too young, or the colors are garish, or the hands are in the wrong place. This is normal.
Practice: make 3–4 variants with different wording. Don't be afraid to add clarifications: "soft pastel tones," "no text," "realistic style." If the neural network draws something gloomy, write "light, cozy atmosphere."
A typical mistake is asking for a "beautiful card with a caption." Models handle Russian text in images poorly, and you get gibberish. Better to generate a clean background and overlay the text manually in any editor (even in Word or free Canva).
Step 3. Find warm text — with the help of an AI twin or on your own
If you don't know where to start, the AI twin on asibiont.com can put together a draft greeting, taking your details into account. For example, you say: "Grandma loves cabbage pies, she has a cat named Barsik, she taught me to play chess." The AI will produce 3–4 text variants. Then it's your job: choose, rewrite in your own voice, add something personal.
You can do it without AI too: write down 5 facts about grandma that you remember from childhood and insert them into the greeting. The main thing is not to leave template phrases. Replace "I wish you good health" with "May your knee stop hurting, and may you be able to walk in the park every day."
Step 4. Assemble the card: picture + text + frame
Take the generated image and overlay text on it. The easiest way is in a free editor: Canva, Figma, or even PowerPoint. Choose a serif font (it looks warmer), add a thin frame if needed.
Format: for printing — PDF or PNG in good resolution (at least 2000 pixels on the long side). For sending in a messenger, 1200 pixels is enough. If you want to print it at a photo lab, save as JPEG with 90% quality.
Timelines: image generation — 2–3 minutes, text selection — 10 minutes, layout — 5 minutes. Total about 20 minutes if you don't get stuck on perfection. If you don't like the picture, don't spend an hour redoing it — generate a new variant with a different description.
Where mistakes are most common
- Too general a description. "Beautiful landscape" — you'll get a store-bought card. "The view from her kitchen window: an apple tree, a fence, a cat on the windowsill" — that's already personal.
- Forgetting about the format. The picture may be beautiful, but when printed the edge with the text will be cut off. Leave margins.
- Trusting the neural network's text entirely. AI may insert an inappropriate joke or an overly pompous phrase. Always read it aloud: does it sound like you?
- Trying to do everything in 5 minutes. Realistically — in 20–30, if you don't get distracted.
What if things don't go according to plan
The neural network produces strange pictures. Try another model or change the description: remove complex details (for example, "cross-stitch embroidery"), add "simple subject, clean colors."
No time to print. Send the card in a messenger — grandma will be happy anyway. Many elderly people save such pictures in their gallery and show them to their friends.
Grandma doesn't use a smartphone. Then only printing: the nearest photo kiosk or a printer at friends' place. You can send the electronic version to her children so they can show her.
A non-obvious conclusion: a card is an occasion for conversation
When you make a personalized card, you inevitably recall details: her favorite apple variety, her habit of saying "oh, come on." These details can then be used in conversation. Grandma will see that you remember, and that's more valuable than the picture itself.
Another point: don't try to make it perfect. A uneven edge or a simple drawing often touches more than a glossy masterpiece. What matters in a card is not aesthetics, but attention.
If you want to simplify the routine, the AI twin on asibiont.com can generate a picture from a description, put together a draft greeting, and even remind you of the date if you ask. But printing, signing, and personal words are up to you. Start simple: describe a scene that you associate with grandma, and see what happens.
And if tomorrow is a colleague's or friend's birthday — the principle is the same: a specific detail, a generated picture, and a couple of warm words written by hand.