37-frame animatic board: $0.93 to $6.94 by Sume image row

A 37-frame storyboard of stills costs $5.55 on Nano Banana 2.1 at 2K, $1.62 on Seedream 5 Lite, and $2.06 on GPT Image 2.5 at medium quality and 2K, using Sume's published per-image prices as of 2026-10-09. The cheapest row, grok-image, brings the same board to $0.93, and Nano Banana Pro at 2K is the top of the range at $6.94. A mixed plan, with 37 low-quality drafts and six Pro finals, comes to $2.04.

A board is mostly a counting problem: frames times price, then a few redraws. This page does the counting so you can set a wallet limit before you start.

Cost of the whole board

Each line multiplies the price per image by 37 frames. No extras are included: no references, no redraws, and no upscaling.

Add redraws and a hero pass

Storyboards are rarely one pass. If a quarter of the frames get redrawn once, add 10 images (a quarter of 37 rounded up). At Seedream 5 Lite prices that adds $0.44. At Nano Banana 2.1 2K it adds $1.50.

A cheaper pattern is drafts first, finals after. Thirty-seven GPT Image 2.5 low drafts at $0.02475 cost $0.92. Pick the six frames that carry the story and regenerate them on Nano Banana Pro at 2K, $0.1875 each, for $1.1250, so $1.12. The board plus finals then costs $2.04, which is less than a single pass of 37 frames at Nano Banana 2.1 2K.

Keep frames consistent

The Image API accepts input_references on models that list them, so you can feed a character frame into later frames. The reference count limit depends on the model: GPT Image 2.5 takes up to 16, and Ideogram 4.5 takes up to 5 in total when editing. Sume's docs state that a request which sets a parameter a model does not list gets 400 unsupported_parameter, so check the model's descriptors before you add references.

Use one aspect_ratio across the board. For a film frame, 16:9 and 21:9 are both in Sume's normalized list, but a model accepts only the values its catalog entry lists.

A worked pass for a 37-frame board

Here is a concrete run order. Generate all 37 frames on GPT Image 2.5 at low quality, which costs $0.92 (37 times $0.02475, $0.91575 before rounding). Review them as a strip and mark the frames that need a redraw. Redraw nine of them on the same row for $0.22. Pick six hero frames and regenerate them on Nano Banana Pro at 2K for $1.13. The running total is about $2.26, and every number above comes straight from the Sume price list rather than from a measurement of your prompts. Your redraw rate will differ, so track the real figure from usage.cost and adjust the plan after the first ten frames.

Set the wallet limit

Sume bills only completed generations. To cap the exposure, make n match the frames you want per call and read usage.cost in each response; the Image API returns the billed amount per call. Slow configurations can return 202 with a job envelope instead of the image, so poll the job rather than resubmitting, because a resubmitted request is a new generation. The jobs and results guide shows the polling endpoints.