Ideogram 4.5 ad text: cost of low drafts then one high final

If you iterate on ad headlines with Ideogram 4.5, the cheapest plan is to draft every concept at low quality and re-run only the winners at high. On Sume, six low drafts cost $0.225 and two high finals cost $0.55, so the whole round is $0.775. Six concepts straight at high would cost $1.65.

The arithmetic below uses Sume's Ideogram 4.5 prices, $0.0375 at low, $0.075 at medium and $0.275 at high per image, which are the list prices in the image docs with Sume's 25 percent applied.

What is the draft-then-final plan?

Ideogram 4.5 generates from text when you send no input_references, and the price depends on quality only. That makes a two-pass round easy to cost: pass one is many images at the cheap quality, pass two is a few images at the expensive one.

A final is a new generation, not an upscale of the draft. Re-run the same prompt text at high, and compare. The docs make no promise that a high render will keep the exact composition of a low draft, so budget for the final as a fresh image.

What does one round cost compared with the alternatives?

Take six headline concepts and two finals as the example. The table prices three ways to run the same round at the Sume per-image prices.

How does the saving grow over a week of rounds?

The saving is the gap between running every concept at high and running only finalists there. For one round it is $0.875. Run one round on each of five working days and the draft plan costs $3.875 against $8.25 at high throughout, a difference of $4.375.

The saving shrinks as more drafts survive. If four of six survive, the plan costs $0.225 plus $1.10, or $1.325, which is still below $1.65 but no longer a big gap. The plan pays off when you keep few.

When does the draft plan stop saving money?

The plan costs six low drafts, $0.225, plus every survivor at high, $0.275 each. Running all six at high is $1.65. The draft plan stays cheaper as long as the drafts you pay for are smaller than the high renders you skip, which holds up to five survivors.

At six survivors you have paid for the drafts and then re-run everything, so the plan costs $1.875, which is $0.225 more than skipping the drafts. The table shows the full range for one round of six concepts.

What should I check before sending the batch?

Each submit reserves its estimate before it starts. A high request for two images needs $0.55 of spendable balance, and a low request for six needs $0.225. If the balance is short, the submit returns 402 insufficient_credits and nothing runs. Failed generations are not billed.

To bound the damage of a script that loops, send each paid submit with an Idempotency-Key, and read GET /v1/usage after the round. The ledger shows reserved, captured and refunded rows, so you can compare the real spend with the $0.775 above.