Child's drawing to a picture-book page with GPT Image 2.5

To turn a child's drawing into a picture-book page, upload the scan to a public HTTPS URL, send it as one input_references entry to GPT Image 2.5, and ask for a 4:3 page in your prompt and aspect_ratio. Say what to keep: the creature's odd proportions, the colors, the crooked smile.

The common failure is over-polish. If the prompt says only "make it professional", the model smooths away the thing that made the drawing special. Name the features to preserve.

Write the prompt as keep and change lists

Split the prompt in two. Keep: the big round head, the three legs, the orange and purple palette, the wobbly line. Change: add a soft watercolor paper texture, a simple meadow background and room at the top for text. Then state that the picture must stay recognisable to the child who drew it.

Phone scans are the usual input. Crop the page edge and flatten shadows first, and use a JPEG or PNG. Sume only reads references from public HTTPS URLs, so a private link or a local path fails with 400 input_media_unreachable.

Request

Reference URLs must be public HTTPS, and Sume answers 400 input_media_unreachable when it cannot download one. A file on your laptop needs a public upload first, for example through the assets upload flow.

import os, requests
H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}

def generate(body):
    r = requests.post("https://api.sume.com/v1/images", headers=H, json=body, timeout=60)
    if r.status_code != 200:  # 202 = still running, read data.status_url
        raise SystemExit(f"{r.status_code}: {r.text[:300]}")
    return r.json()["data"][0]["url"]
from io import BytesIO
from PIL import Image

url = generate({
    "model": "openai/gpt-image-2.5",
    "prompt": "Make a picture-book page from this child's drawing. Keep the big round head, three legs, orange and purple colors and the wobbly line. Add watercolor paper texture and a simple meadow, with empty space at the top for text.",
    "input_references": [{"type": "image_url", "image_url": {"url": "https://example.com/drawing-scan.jpg"}}],
    "aspect_ratio": "4:3",
    "quality": "medium",
    "output_format": "png",
})
Image.open(BytesIO(requests.get(url, timeout=60).content)).save("page-1.png")
print("saved page-1.png")

Cost per page

GPT Image 2.5 is billed on tokens. The fal pages list $30 per million output image tokens, $8 per million image input tokens and $5 per million text input tokens. Sume bills the provider list price times 1.25. The table is output-only, so reference and prompt tokens add a little on top of it.

A 4:3 page at a 1024x768 size is a useful draft. The table prices that size.

Keep the character across pages

For page two, pass the original drawing and the approved page one as two references, and name which image holds the character. See the storybook post for the consistency workflow.

If the call returns 202

POST /v1/images waits up to 30 seconds and returns 200 with the images. A slow job falls back to a 202 job envelope, and you read the images from GET /v1/jobs/{id}/result. The code above exits on any non-200 so you notice, and a failed synchronous job returns 502 and is not billed.

Ask the child, then check the page

The goal is a page the child still recognises as theirs. A short check against the original catches the common drift before you build a whole book.

• Read the page next to the original: is every feature you listed still there?
• If the style drifted, shorten the prompt and name one art style only.
• Save a PNG of each approved page so the next page can reference it.