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.

5 min readSume
All posts

An AI logo generator drafts logo concepts from a prompt that names the brand, the symbol, the style and two or three colors, and one call can return several options to choose from. On Sume, what comes back is a raster image, a PNG, JPEG or WebP rather than an SVG, and any lettering can come out misspelled. Pick a concept, cut it out to a transparent PNG for websites and profiles, and have the final mark redrawn as a vector in a design tool when you need one that scales for print or signage.

Sume facts come from the Image API docs and the OpenAPI document behind the Sume API reference, read on 2026-09-28.

What should a logo prompt include?

Say what the mark is, how it looks, and what it sits on. A prompt that covers these leaves the model less to guess:

  • The brand name in quotes, spelled exactly, if it belongs in the mark. Proofread every letter of the result.
  • One symbol or idea, such as a leaf, a mountain or an initial.
  • A style: flat, geometric, line art, badge or mascot.
  • Two or three named colors.
  • A plain white background with no shadow, texture or mockup, so the mark cuts out cleanly and is easy to redraw later.
  • n for several concepts in one call: the current catalog allows up to 4 on ChatGPT Image 2.5 (openai/gpt-image-2.5).
  • quality: ChatGPT Image 2.5 takes auto, low, medium, high, xhigh or max, and defaults to high. AI image with text generator covers lettering and which setting to raise.
  • A 202 instead of 200 means the concepts are still being made: poll GET /v1/jobs/{id}/status, then read GET /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": "Flat minimal logo for a bakery called \"Acme Bakery\": an oak leaf shaped like a wheat ear, warm brown and cream, centered on a plain white background, no shadow",
    "n": 4,
    "quality": "high",
    "output_format": "png"
  }'

Can an AI logo generator make an SVG?

Not on Sume. Each image model in today's catalog lists PNG, JPEG and WebP as output formats, or fewer. svg appears among the schema's output_format values, but no model lists it, and a model only accepts the values its catalog lists: current code refuses svg with 400 invalid_request and names the formats the model does support. For a logo that scales to any size, trace or redraw the chosen PNG as a vector in a design tool.

From Image API and the Sume API reference, read 2026-09-28. The svg row is current code.
You wantWhat to sendWhat you get
PNG, JPEG or WebPoutput_format: "png", "jpeg" or "webp" on POST /v1/imagesThat raster format, at a signed Sume-hosted URL
SVGoutput_format: "svg"400 invalid_request: no model lists svg
A transparent backgroundThe chosen image's URL to POST /v1/rmbg-1.0/removeA PNG with alpha
A bigger fileThe image's URL to POST /v1/image-upscale-1.0/upscaleThe image enlarged by upscale_factor 1–4

How do I get a logo with a transparent background?

Cut out the concept you picked: send its URL as image_url to POST /v1/rmbg-1.0/remove, and the completed job returns a PNG with alpha. Background removal separates the mark from what's behind it, so check white or light parts inside the logo, such as letter counters, at full size. Keep the file as a PNG, because a JPG has no alpha channel. AI image generator with a transparent background walks through both calls.

Can I make a logo from a sketch or an old logo?

Yes. Put the sketch or the old logo at a public HTTPS URL and send it in input_references, with a prompt that says what to keep and what to change. ChatGPT Image 2.5 takes up to 16 reference images. The model draws a new image, so shapes and letters can shift; compare the result with your original. Sketch to image covers turning drawings into finished art.

How much does it cost, and what are the limits?

Each concept is priced per model. ChatGPT Image 2.5 estimates each request's price from its size and quality, so read usage.cost in the response: it is the USD amount billed. Failed generations are not billed. A cutout is $0.0225 per image on API pricing, plus a 5.5% agent fee by default.

  • No vector output: plan a redraw if you need SVG, PDF or another vector format.
  • Spelling is not guaranteed. If the name must be exact, set it yourself in a font next to a symbol-only mark.
  • Result URLs from POST /v1/images are signed, so download the concepts you keep.
  • Once the mark is final, Logo animation API covers turning it into a short ident.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume