AI image generator for presentations: 16:9 slide images

To make AI images for a presentation, generate them at the slide's shape: new PowerPoint presentations default to 16:9 widescreen, so ask for 16:9 images, such as 2560 × 1440 or 3840 × 2160. Keep the deck consistent by reusing one approved image as a reference and repeating the same style words in every prompt, leave empty space where the slide's words go, and type those words in PowerPoint instead of asking the model to draw them. With Sume, that is POST /v1/images, and ChatGPT Image 2.5 takes those exact pixel sizes.

PowerPoint facts come from Microsoft Support's page on changing slide size, and Sume facts from the Image API docs, both read on 2026-09-28. For a talking presenter on each slide instead of a still, see talking avatar for PowerPoint presentations.

What size should AI images for PowerPoint be?

Match the slide. Microsoft says the 16:9 widescreen setting is the default value for new presentations; the Slide Size menu's other preset is Standard (4:3), and Custom Slide Size covers the rest. Widescreen slides measure 13.333 by 7.5 inches, and PowerPoint can't give one slide a different size from the others, so one shape fits every full-slide image in a deck.

Inches don't set pixels, so pick the pixel size for the screen you present on. ChatGPT Image 2.5 takes these as exact custom sizes, with both edges in multiples of 16; 1080 isn't one, so the 1080-pixel rows need a small crop. What size is 16:9 in pixels? lists every 16:9 size it accepts.

How do I keep every slide image in the same style?

Reuse one approved image as a reference, and repeat one short style line in every prompt:

• Make the first image with n set to 2, 3 or 4 to compare options; ChatGPT Image 2.5 returns up to 4 per call today. Pick one.
• Download it and host it at a public HTTPS URL, because result URLs are signed.
• Send it in input_references with each new prompt, and paste the same style line each time, such as "flat vector illustration, navy and coral palette".
• Change only what the slide is about. No setting guarantees a matching style, so line the images up side by side before you build the deck.

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": "Same style as the reference: flat vector illustration, navy and coral palette. A team planting a tree on the right third; calm empty background on the left for a title. No text.",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/deck-style.png" } }
    ],
    "image_size": { "width": 2560, "height": 1440 },
    "n": 2
  }'

How do I leave room for the words on a slide?

Say where the empty space goes, as you would brief a designer: "subject on the right third, calm empty background on the left for a title, no text". Then type the title and bullets in PowerPoint, where they stay editable; words drawn into a picture can't be edited or corrected later. If you do want words inside an image, AI image with text generator covers how to prompt for them and check them.

Can I put an AI illustration on a colored slide?

Yes, once it is cut out. On Sume, a transparent background comes from background removal, a second step after the image is made: ask for the illustration on a plain background, then send the image's URL to POST /v1/rmbg-1.0/remove, which returns a PNG with an alpha channel. It costs $0.0225 per image on API pricing, plus a 5.5% agent fee by default. AI images with a transparent background walks through the two steps.

What does it cost, and what are the limits?

• Each model has a per-image rate in the pricing line of GET /v1/images/models/{model_id}/endpoints, plus a 5.5% agent fee by default. On ChatGPT Image 2.5, that price is estimated from tokens, so the image's size and quality and the reference images you send all move it. usage.cost in the response is the USD amount billed, and a failed generation is not billed.
• The docs publish exact-size rules only for the ChatGPT Image models. With other models, send a ratio such as 16:9 or 4:3 and check the pixel size you get back.
• Result URLs are signed, so download the images before you build the deck.
• The images are stills, and a large or high-quality request can take longer than the call's 30-second wait: then it returns 202 and a job to poll.