AI sticker generator: from a prompt or photo to a PNG cutout

An AI sticker generator works in two steps: generate the design as one flat illustration with a bold outline on a plain background, then cut it out with background removal, which leaves the design on a transparent PNG. For printed stickers, work out the pixels first (a 3-inch sticker at 300 DPI is 900 pixels wide); to turn a photo into a sticker, send the photo as a reference image and ask for a sticker version of it. On Sume that is POST /v1/images, then POST /v1/rmbg-1.0/remove.

Sume facts come from the Image API docs and the OpenAPI document behind the Sume API reference, read on 2026-09-28. Inches × DPI is plain arithmetic; AI image for print explains it for any print.

How do I write a sticker prompt?

Describe the subject, then the sticker style. Unlike a coloring page, a sticker keeps its color and loses its background, so the prompt should make that background easy to remove:

• One subject per image, whole and centered, with empty space around it.
• Flat colors and a thick outline around the whole design.
• A plain background in a color the design doesn't use.
• For a white border like a die-cut sticker, describe it: “a thick white border around the whole character”. It is part of the drawing, not a setting, so check that it goes all the way round.
• Few or no words, and check their spelling. AI image with text generator covers lettering.
• Several variations per call with n: the current catalog allows up to 4 on ChatGPT Image 2.5 (openai/gpt-image-2.5).

How do I turn a photo into a sticker?

Send the photo in input_references as a public HTTPS URL, with a prompt that names the change: “turn this dog into a flat cartoon sticker with a thick white border, plain gray background”. ChatGPT Image 2.5 takes up to 16 reference images; models whose reference range is 0 to 0 are text-to-image only and reject the photo.

The model draws a new image from the photo, so likeness, markings and small details can change. Compare the sticker with the photo before you print it.

What size should a sticker image be?

Pixels are inches × DPI. If your printer asks for 300 DPI, each inch of sticker needs 300 pixels. Size, bleed and how the cut line is supplied are your print shop's rules, so check its file requirements before you order.

ChatGPT Image 2.5 takes an exact custom image_size when both edges are multiples of 16, the longest edge is at most 3840 and the total is 655,360 to 8,294,400 pixels. The smallest custom square it accepts is 816 × 816, so small stickers are generated a little large; the largest is 2880 × 2880, 9.6 inches at 300 DPI. For bigger stickers, enlarge the image with Image Upscale 1.0 by a factor from 1 to 4. For each size below, ask for the nearest legal square at or above the target:

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": "Sticker of a smiling cactus in sunglasses, flat colors, thick dark outline, thick white border around the whole design, plain gray background",
    "image_size": { "width": 912, "height": 912 },
    "n": 4,
    "output_format": "png"
  }'

How do I cut the sticker out?

If the generation answered 202 instead of 200, the images are still being made: poll GET /v1/jobs/{id}/status, then read GET /v1/jobs/{id}/result. Then send the chosen image's URL as image_url to POST /v1/rmbg-1.0/remove. The job runs async by default; when it is completed, the result is PNG artifacts with alpha. Remove background API covers the call.

• Upscale first, cut out last. The API reference doesn't describe how the upscaler treats an alpha channel, so enlarge the opaque image and cut out the larger file.
• Check the artifact's width and height, when reported, before you send the file to print.
• Sume returns pixels, not a vector cut path. Ask your print shop how it wants the cut line.

How much does an AI sticker cost?

The image is priced per model. ChatGPT Image 2.5 estimates each request's price from its size and quality (in current code its catalog price line is for one 1024 × 1024 image at high), so read usage.cost in the response: it is the USD amount billed. Failed generations are not billed. API pricing lists the cutout at $0.0225 per image and an upscale at $0.20 per image, plus a 5.5% agent fee by default.