Photo to line drawing on a transparent background with GPT Image 2.5

To turn a photo into a line drawing with a transparent background on Sume, send the photo as one input_references entry to openai/gpt-image-2.5 with background: "transparent", output_format: "png" and aspect_ratio: "auto", then open the result in Pillow and confirm the corners have an alpha of 0. Sume's Image API docs list background for ChatGPT Image 2.5 only, so this is the row to pick for a cut-out line drawing.

Which request settings matter?

OpenAI's image generation guide says to set background: "transparent" and use png or webp as the output format. Sume's page repeats the field values auto, transparent and opaque. For an edit, add aspect_ratio: "auto" so the drawing keeps the photo's shape; Sume's docs say that omitting the field is not the same as sending auto.

What does the call look like?

The prompt does the style work. Ask for clean black outlines, no shading, no fill, and no background. The script below sends the request and checks the alpha channel of the first result; a wholly opaque file means the model ignored the transparency request and you should retry before batching.

import io
import os
import requests
from PIL import Image

key = os.environ["SUME_API_KEY"]
body = {
    "model": "openai/gpt-image-2.5",
    "prompt": "Clean black line drawing of this photo, outlines only, "
              "no shading, no fill, no background",
    "input_references": [{"type": "image_url",
        "image_url": {"url": "https://example.com/photo.jpg"}}],
    "background": "transparent",
    "output_format": "png",
    "aspect_ratio": "auto",
    "quality": "medium",
}
r = requests.post("https://api.sume.com/v1/images",
    headers={"Authorization": f"Bearer {key}"}, json=body, timeout=60)
if r.status_code != 200:
    raise SystemExit(f"{r.status_code}: {r.text[:200]}")
img = Image.open(io.BytesIO(requests.get(r.json()["data"][0]["url"]).content))
img = img.convert("RGBA")
print("corner alpha:", img.getpixel((0, 0))[3])

What if the request returns 202?

Sume waits up to 30 seconds on POST /v1/images. A slow request returns 202 with a job envelope instead of the image, and the script above exits on that. Read the result from the job endpoints described in Jobs and results, or send mode: "async" from the start for batches.

How do I know the drawing is clean?

Check the file, not the preview. A viewer often shows a checkerboard or white page whether or not the file has an alpha channel. Convert to RGBA in Pillow and read the corner pixel: an alpha of 0 means the background is transparent, and 255 means the model returned a flat image. Also open the PNG on a dark background, because stray light-gray halo pixels around the lines only show up there.

Sume bills image generation all or nothing, according to its docs: a completed generation is charged in full and a failed or cancelled one is not. A bad drawing still costs you the render, so test one photo at low before you queue a folder.

Should I use Flare or Sunburst?

Both ids take the same fields on Sume. OpenAI's guide recommends Sunburst for workflows where editing precision matters most, which is the case here because you want the photo's contours kept. Try both on one photo; the pair share the same Sume request shape, so only model changes.