AI invitation generator: card art with room for the details
Generate invitation art at card shape, such as 1600 × 2240 for a 5 × 7-inch card at 320 DPI, then type the names, date, and address yourself.

To make an invitation with AI, generate the artwork at the card's shape with a clear panel for the words, then type the names, date, time, and address yourself, because an image model can misspell details that must be exact. On Sume, ChatGPT Image 2.5 takes 1600 × 2240 pixels, a 5 × 7-inch card at 320 pixels per inch, and can build the design around a photo of the couple, the birthday child, or the venue that you send as a reference.
Sume facts come from the Image API docs and the request schemas in the Sume API reference, read on 2026-09-28. Card sizes and pixels per inch are plain arithmetic.
What size should an AI invitation be?
Pick the card size first, then generate at its exact shape so nothing gets cropped. 320 pixels per inch is a handy target for ChatGPT Image 2.5: 320 is a multiple of 16, so any card size in whole or quarter inches gives edges in multiples of 16, one of the rules for its custom sizes. It is also above 300, if your print shop asks for 300 DPI.
| Card | Shape | `image_size` at 320 PPI |
|---|---|---|
| 5 × 7 in | 5:7 | 1600 × 2240 |
| 4 × 6 in | 2:3 | 1280 × 1920 |
| 4.25 × 5.5 in | 17:22 | 1360 × 1760 |
| 5 × 5 in | 1:1 | 1600 × 1600 |
How do I generate the invitation art?
Send POST /v1/images with openai/gpt-image-2.5 and the card size as image_size, and describe the art and the space for the words:
- The occasion and mood: "a garden wedding, eucalyptus and white roses, soft watercolor", or "a dinosaur fifth-birthday party, bright cartoon style".
- The empty space: "a plain cream panel in the center for text, with the flowers framing the edges".
- "No text, no letters", so the panel stays empty.
nfor up to 4 versions per call, this model's ceiling today. Each completed image is billed.
curl -X POST "https://api.sume.com/v1/images" \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-image-2.5",
"prompt": "Wedding invitation artwork: eucalyptus and white roses framing the edges, soft watercolor, a plain cream panel in the center for text. No text, no letters.",
"image_size": { "width": 1600, "height": 2240 },
"n": 4
}'Should the AI write the names and date?
No, type them: generated lettering can come out misspelled, and today the file comes back as a flat PNG, JPEG, or WebP with no text layer to fix. AI image with text explains why, with prompts for one short drawn word such as "Party!". In a layout tool you can check every name, date, and address, choose the typeface, and reuse the same art for a save-the-date or an RSVP card.
Can I put a photo of us or the venue on the invitation?
Yes, as a reference. Put the photo at a public HTTPS URL and send it in input_references with a prompt that says how to use it, such as "the couple from the photo as a soft watercolor illustration under an arch of white roses". Keep image_size in the request: it wins over aspect_ratio, so the card keeps its 5 × 7 shape, while aspect_ratio: "auto" would follow the photo's shape instead. AI greeting card generator shows the full request.
The model redraws the people, so a likeness isn't guaranteed. Check faces before you print, or place the real photo in the layout tool.
Can I make a digital invitation to send by text or email?
Yes. For an invitation people open on a phone, generate the art with aspect_ratio: "9:16", the tall shape of a phone screen, which ChatGPT Image 2.5 lists today, instead of a print size. Add the words in a design app as you would for print and export a JPEG or PNG. Keep the date and place large, since the whole card shows at phone size.
What does it cost, and what are the limits?
- Each completed image is billed, and a failed or canceled one is not. ChatGPT Image 2.5 estimates each request's price from its size and quality, and
usage.costin the response is the USD amount billed. - High
quality, ChatGPT Image 2.5's default, and a largenare two of the slow settings most likely to return202with a job instead of the images. Poll the job's status until it is terminal, then read its result. - Reference photos must be public HTTPS URLs; localhost, private-network, and non-HTTPS URLs are rejected.
- Results are Sume-hosted, signed URLs, so download the art before you lay it out.
Sources
Related posts
More in Use cases
- AI jewelry product photography from one real photo
AI jewelry product photography works best as an edit: send a real photo of the piece, change only backdrop and light, then check every stone.
- AI video for jewellery: animate a real jewelry photo
Make an AI jewellery video from a real photo: pin it as the first frame, ask for one slow move, keep it short, and check stones frame by frame.
- AI lofi video generator: a looping scene plus a lo-fi track
An AI lofi video is a short looping scene under a lo-fi track. Generate the loop and the music, then repeat both for up to 30 minutes in one render.
- AI logo generator: PNG concepts from a prompt, not SVG
An AI logo generator drafts logo concepts from a prompt. On Sume they come back as PNG, JPEG or WebP, not SVG; cut one out for a transparent PNG.
Written by Sume