Replace headline text on a finished ad with Ideogram 4.5 on Sume

Send the finished ad as the first reference to ideogram/ideogram-v4.5 on Sume, name the new headline, omit aspect_ratio. List cost $0.03-$0.22.

4 min readSume
All posts

To change the headline on a finished ad with Ideogram 4.5 on Sume, send the ad as the first input_references item, describe the new text in the prompt, and leave out aspect_ratio so the result keeps the ad's shape. The Higgsfield changelog (read 2026-10-05) lists reference edits and text replacement for Ideogram 4.5 as of Sep 30, which is the vendor claim this recipe puts to the test.

How Sume's Ideogram 4.5 route works

From the Image API docs: with no references, ideogram/ideogram-v4.5 generates from text. With references, it edits the first image and uses up to 4 more as references, 5 in total.

Ideogram 4.5 on Sume, Sume docs, read 2026-10-05
SettingValue
Model idideogram/ideogram-v4.5
First referenceThe source image to edit
Extra referencesUp to 4 more (5 total)
qualitylow, medium (default) or high
resolution1K or 2K
aspect_ratioOmit on an edit to keep the source shape
Price per image (provider list)$0.03 low, $0.06 medium, $0.22 high, same at all sizes

The request

Quote the old and new headline in the prompt and say everything else must stay. Use a small test image first. The call below uses medium quality.

import asyncio, os, httpx

async def main():
    body = {
        "model": "ideogram/ideogram-v4.5",
        "prompt": "Replace the headline 'SUMMER SALE' with 'AUTUMN SALE'. Keep the layout, colors and photo unchanged.",
        "quality": "medium",
        "input_references": [{"type": "image_url", "image_url": {"url": "https://example.com/ad.png"}}],
    }
    headers = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
    async with httpx.AsyncClient(timeout=60) as c:
        r = await c.post("https://api.sume.com/v1/images", headers=headers, json=body)
    print(r.status_code, r.json())

asyncio.run(main())

Cost and what to check

Sume bills provider list x 1.25, so the three list prices become $0.0375, $0.075 and $0.275 per image before cent rounding. The endpoints pricing line has the billed figure. A draft-then-final run (two low drafts, one high final) is $0.03 + $0.03 + $0.22 = $0.28 list, or $0.35 after x 1.25.

  • Spelling: read every letter of the new headline in the output. We make no claim about how reliably the model renders text.
  • Untouched areas: compare logo, product and background against the original. A full regeneration can alter them.
  • Size: 1K and 2K cost the same, so test at 1K and run the final at 2K.
  • Mask: mask_url works only on ChatGPT Image 2.5 on Sume, so there is no mask for this route.

If the text drifts

Shorten the headline, put it in quotes, and rerun at the next quality step. If the layout itself moves, try the masked edit on ChatGPT Image 2.5, which accepts a mask_url and up to 16 references. For the shape rule, see the aspect ratio note on Ideogram edits.

Batch the same edit

If you have one ad in ten sizes, run the headline edit on each size rather than resizing the edited result, since an edit without aspect_ratio keeps the shape of its source. Run the first at low quality to check the wording, then the rest at the quality your review passes. At $0.03 list for low, ten checks cost $0.30 at list, or $0.375 after the x 1.25 on Sume (before cent rounding).

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume