First-frame stills for 500 SKUs: five image models priced

If your video model is image-to-video, every SKU needs a good first frame before it needs motion. Generating that frame for 500 SKUs costs between $32.95 and $131.95 on Sume depending on the image model, at high quality, text-to-image (prices read 2026-10-05 from the Sume pricing code, billable at provider price x 1.25). That is a four-fold spread for the same count of stills, so the model choice matters more than the count. Stills are also the cheapest stage of the pipeline, a small fraction of what motion costs, which makes them the right place to be picky: a bad first frame wastes the clip that follows it, and a clip costs roughly 10 to 90 times as much as a $0.0659 frame (from $0.625 for a 5-second Wan 3.0 clip to $5.67 for 15 seconds of Seedance 2).

The price grid

One image per SKU, 500 SKUs, no retries. Multiply by your expected attempts per SKU for the real bill.

How to read it

The 1,000-attempt column covers two tries per SKU, which is a realistic planning number when a first frame has to show a label or logo correctly. Even doubled, the cheapest model costs $65.90 for 500 SKUs, about the price of 12 Seedance 2 clips of 15 seconds at 720p ($5.67 each).

Where you already have a product photo, an edit with one reference is cheaper: gpt-image-2.5 costs $0.0094 low, $0.0209 medium and $0.0835 high per edit, and accepts up to 16 references. Edits at high quality cost more than text-to-image at high ($0.0835 against $0.0659), so use edits when the photo must be preserved, not to save money.

• Fix the aspect ratio before generating: stills for a 9:16 ad should be made at 9:16, not cropped later.
• Run 20 SKUs on two models and compare before committing the other 480.
• Spend high quality only on SKUs with text on the pack.
• Keep low-quality edits for the draft gate, then re-run approved SKUs at high.

For a cheap approval step before motion, see the $0.0094 draft gate. For what the motion stage adds, see the tiered 300-SKU budget.