AI T-shirt design generator: art on a transparent PNG
Generate T-shirt art on a plain background, upscale it to your printer's pixel size, then cut it out so the design sits on a transparent PNG.

An AI T-shirt design generator makes the print artwork itself, not a photo of a shirt. Generate the design on a plain background, one design per image, enlarge it to the pixel size your print shop's template asks for, then cut it out so the shirt color shows around it: the cutout is a PNG with a transparent background. On Sume that is POST /v1/images, POST /v1/image-upscale-1.0/upscale and POST /v1/rmbg-1.0/remove, in that order. Seeing the design on a shirt is a separate mockup step.
Sume facts come from the Image API docs and the OpenAPI document behind the Sume API reference, read on 2026-09-28. Pixels from inches × DPI is plain arithmetic.
How do I prompt for T-shirt art?
Describe the graphic, not the shirt, so the image is the artwork alone:
- One design per image, centered, with nothing cropped at the edges.
- A style and a short list of colors: “vintage badge, three colors, thick lines”.
- A plain background in a color the design doesn't use, so the cutout has a clean edge.
- Short, exact words in quotes. Proofread every letter, or set the words yourself in a design tool; AI image with text generator covers lettering and the
qualitysetting. - Several designs per call with
n: the current catalog allows up to 4 on ChatGPT Image 2.5 (openai/gpt-image-2.5). - A
202instead of200means the designs are still being made; the docs name highqualityand a largenamong the settings most likely to cause it. PollGET /v1/jobs/{id}/status, then readGET /v1/jobs/{id}/result.
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": "T-shirt graphic: a retro mountain sunset badge, three flat colors, thick lines, centered on a plain light gray background",
"image_size": { "width": 2448, "height": 3264 },
"n": 4,
"output_format": "png"
}'What size should a T-shirt design be?
Your print shop sets it: a print area in inches and a resolution, such as 300 DPI. Multiply one by the other for the pixels; AI image for print explains the arithmetic for any print. ChatGPT Image 2.5 takes an exact custom image_size with both edges in multiples of 16, a longest edge of at most 3840 and at most 8,294,400 pixels, so generate the largest frame in your design's shape, then upscale by a factor from 1 to 4 if the print area needs more. Decimal factors are allowed, and the docs don't say how the result is rounded, so check the returned size.
| Print area (example) | Pixels at 300 DPI | Largest frame | Upscale factor |
|---|---|---|---|
| 8 × 10 in | 2400 × 3000 | 2560 × 3200 (4:5) | None needed |
| 10 × 10 in | 3000 × 3000 | 2880 × 2880 (1:1) | 1.1 (about 3168 × 3168) |
| 12 × 16 in | 3600 × 4800 | 2448 × 3264 (3:4) | 1.5 (about 3672 × 4896) |
How do I get a transparent PNG for printing?
Upscale first, cut out last. The API reference doesn't describe how the upscaler treats an alpha channel, so enlarge the opaque image and make the transparency in the final step:
- Send the chosen design's URL as
image_urltoPOST /v1/image-upscale-1.0/upscalewith yourupscale_factor. It returns PNG by default. - Send the upscaled image's URL to
POST /v1/rmbg-1.0/remove. Both run as jobs,asyncby default; the completed cutout is a PNG with alpha. - Check the artifact's
widthandheight, when reported, against your print shop's size before you send the file. - Keep the file as a PNG; a JPG can't store the transparency.
Can I see the design on a shirt first?
Yes, as a separate image: send the finished design and a photo of a blank shirt as references and describe where the design goes. AI mockup generator API covers that request. The mockup is for review; print from the cutout.
How much does it cost, and what are the limits?
Each design 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 an upscale at $0.20 per image and a cutout at $0.0225 per image, plus a 5.5% agent fee by default.
- Raster files only: PNG, JPEG or WebP, not SVG.
- Color, file-size and resolution rules belong to your print shop; check its file requirements.
- Result URLs from
POST /v1/imagesare signed, so download the designs you keep.
Sources
Related posts
More in Use cases
- AI tattoo generator: design ideas and a preview on your skin
Ask an AI for tattoo designs as black linework on white, preview one on a photo of your arm by sending both images, then take it to a tattoo artist.
- AI training video generator: from written steps to video
An AI training video generator turns written steps into presenter videos. Build one module from short avatar clips: one clip per step, captioned, joined.
- Arcads alternatives: AI UGC ad tools by plan and API
Arcads alternatives for AI-actor UGC ads: Creatify, HeyGen, Higgsfield and Sume compared on presenters, API, billing and free options, read 2026-09-28.
- Arcads vs Creatify: AI actor ads or URL-to-video ads
Arcads builds ads from your script and 1,000+ AI actors on monthly plan credits. Creatify turns a product URL into a video ad; API plans start at $99.
Written by Sume