Ideogram 4.5 low, medium or high: a text grid to pick the tier

To choose between Ideogram 4.5's low, medium and high quality, generate the same headline at all three on Sume, look at the three files at the size they will be shown, and keep the cheapest tier whose lettering you can read without squinting. The tier changes the price by a factor of about seven, and Ideogram's own model page says the size does not change it, so the grid below is the whole experiment.

Ideogram's page on fal describes the model as generating "high-quality images, posters, and logos with Ideogram 4.5, with accurate text rendering", and lists three tiers at $0.03, $0.06 and $0.22 per image (read 2026-10-06, fal's Ideogram V4.5 page). On Sume the model is ideogram/ideogram-v4.5. The Image API docs give the same three list prices, say quality is low, medium or high with medium as the default, and say resolution is 1K or 2K. Sume bills list price times 1.25, so the figures you will see in usage.cost are higher than the vendor list.

What each tier costs at 100 images

Multiply the list price by 1.25 to get the billed amount per image, then by your image count. The endpoint's pricing line is the authority, so read it for your model before you budget; the table is the arithmetic from the published list.

The grid script

The script sends one prompt three times, once per tier, and prints the billed cost beside each URL. Put the exact headline in quotes in the prompt; Ideogram's value is in-image text, so the headline is the thing you are testing. Open the three URLs side by side and view them at the size your page or feed will use, not at full zoom.

import os
import requests

URL = "https://api.sume.com/v1/images"
HEADERS = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
PROMPT = 'A poster with the headline "OPEN LATE FRIDAYS" in bold type, warm paper texture'

for quality in ("low", "medium", "high"):
    body = {
        "model": "ideogram/ideogram-v4.5",
        "prompt": PROMPT,
        "quality": quality,
        "resolution": "1K",
        "aspect_ratio": "3:4",
    }
    r = requests.post(URL, headers=HEADERS, json=body, timeout=60)
    if r.status_code == 200:
        data = r.json()
        print(quality, data["usage"]["cost"], data["data"][0]["url"])
    else:
        print(quality, r.status_code, "see result_url or error body")

How to read the grid

Judge three things in this order: are all the letters present and spelled right, is the spacing between letters even, and does the layout leave the text where a reader expects it. If low passes all three at the display size, you have saved about 86 percent against high on every image. If low misspells and medium does not, medium is your tier, and the next question is only whether high buys anything you can see.

Keep a short list of prompts that failed at low, and re-run them when a new Ideogram version appears. A tier that was too weak for a long headline may be enough for a two-word title, so test the real copy you will ship, not a stand-in. Run each tier more than once if the first result is borderline, because a single sample can mislead in either direction.

A checklist for judging the three files

Do the judging the same way every time, so the answer does not depend on your mood. Print or view each image at its real display size and go through the list below for each tier.

• Every letter of the quoted headline is present, in order, with no extra characters.
• Letter spacing is even, with no touching or doubled strokes at the display size.
• The text sits where the prompt put it, with margins you can live with.
• Small secondary text, if any, is readable; if it is not, drop it from the art and add it in code.
• The image has no stray text that you did not ask for.

When this grid is the wrong tool

If the text matters legally or financially, such as a price, a date or a phone number, do not rely on any tier. Generate the art without that text and draw the numbers in code, then proofread. If you need to correct one word after the fact, an edit call that quotes the old and new text is cheaper than regenerating at a higher tier; see Fix one typo in an AI-generated image.

The Ideogram model page and the Sume docs agree on prices as of 2026-10-06. Prices change, so rerun the estimate from the live catalog before a large batch; estimate a Sume image batch from the pricing lines shows how.