36 thumbnail test images: list-price totals across 7 image models

A thumbnail test of 12 concepts with 3 variants each is 36 images, and at the list prices on fal's model pages that costs between $1.08 (Ideogram 4.5 at low quality, or FLUX.2 pro at 1024 x 1024) and $5.40 (Nano Banana Pro at its standard 1K rate). Those are fal's list prices as I read them, not Sume's charges: on Sume the amount you pay is the pricing line on each model's endpoint record.

The arithmetic below is only multiplication, so you can redo it with your own counts. The useful decision is not the cheapest model but how many of the 36 you need from a pricey one.

List prices and 36-image totals

Each figure comes from the model's own fal page. Ideogram 4.5 is priced by quality tier, so it appears three times. FLUX.2 pro is priced by megapixel, and the page's example for a 1920 x 1080 output is $0.045. Nano Banana 2 and Pro are priced at a standard rate for 1K, with 4K at double on the Pro page.

A two-stage plan

A cheaper way to spend the same effort is to draft wide and finish narrow. Run all 36 at Ideogram 4.5 low for $1.08, pick the best 3 by eye, then regenerate only those 3 at high for 3 x $0.22 = $0.66. The total is $1.74 at fal list, against $7.92 for rendering all 36 at high. The catch is that quality tiers can change composition, so re-check the three finals instead of assuming they match the drafts.

Sume's docs state that Ideogram 4.5's list price is $0.03, $0.06 or $0.22 per image by quality regardless of size, and that 1K and 2K resolutions cost the same, so the draft-then-finish plan does not need a different size to be cheap.

Read Sume's own price lines

Sume's Image API docs say endpoint pricing lines are the amount charged to your wallet, with Sume's margin already applied, and that a failed or cancelled generation is not billed. The script reads those lines for five ids and multiplies by 36. Run it before you commit a budget, because my fal figures are an input, not a quote.

If a model returns a quality-tiered price, print each line rather than one number, and set quality explicitly in your requests so the tier you priced is the tier you run.

Re-run the totals whenever your counts change. At 20 concepts with 3 variants you have 60 images, and the same table scales linearly: 60 x $0.03 is $1.80 and 60 x $0.15 is $9.00. Nothing in the multiplication is hidden, so the only inputs that can move your budget are the number of images, the tier you choose, and the price your own endpoint record shows.

import os
import requests

H = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
IDS = [
    "ideogram/ideogram-v4.5",
    "bytedance-seed/seedream-5-lite",
    "recraft/recraft-v4",
    "qwen/qwen-image-max",
    "google/nano-banana-2",
]
IMAGES = 36
for model_id in IDS:
    url = "https://api.sume.com/v1/images/models/" + model_id + "/endpoints"
    ep = requests.get(url, headers=H, timeout=30).json()["endpoints"][0]
    for line in ep["pricing"]:
        print(model_id, line["billable"], line["cost_usd"], line["cost_usd"] * IMAGES)