9 colorways from one product photo, 2 takes each: 18 edits priced
18 edits (9 colorways x n=2) from one reference photo: $1.80 on 2.1 at 1K, $1.00 on GPT medium 2K, $6.75 on Pro 4K.

The short answer
Nine colorway mockups from one product photo with two takes each is 18 edited images. On Sume that is $1.80 at Nano Banana 2.1 1K, $2.70 at 2.1 2K, $3.38 at Nano Banana Pro 2K, and $6.75 at Pro 4K. GPT Image 2.5 costs $0.45 at low 1K or $1.00 at medium 2K, as default-options rows.
The photo is sent once per call in input_references. The banana rows are per output image, so the single reference adds nothing there.
Setting up the call
Send the photo as an image_url in input_references, set aspect_ratio to auto so the edit keeps the product's frame, and put the colorway in the prompt. Use n=2 for two takes of the same colorway in one call. Nine calls then cover the set.
If a colorway needs a precise region changed, GPT Image 2.5 accepts an optional mask_url on edits. The banana models do not list it in the docs, so use the prompt there.
| Model and tier | Row per image | Per call (n=2) | 9 calls |
|---|---|---|---|
| Nano Banana 2.1 1K | $0.10 | $0.20 | $1.80 |
| Nano Banana 2.1 2K | $0.15 | $0.30 | $2.70 |
| Nano Banana Pro 2K | $0.1875 | $0.375 | $3.38 |
| Nano Banana Pro 4K | $0.375 | $0.75 | $6.75 |
| GPT Image 2.5 low 1K | $0.02475 | $0.0495 | $0.45 |
| GPT Image 2.5 medium 2K | $0.055625 | $0.11125 | $1.00 |
Which job gets which model
Use the cheapest tier to test colors and the photo's lighting, then promote the approved colorways. Promoting 4 of 9 to Pro 4K with one take each adds 4 x $0.375 = $1.50, on top of $1.80 for the 1K test round: $3.30 in all.
Check the product's text and logo on every output. A mockup that changes a logo is a failed job for you even though Sume bills it as complete.
Billing details
Sume bills only completed generations, and the endpoint pricing already includes the margin. The first response of each model shows usage.cost; compare it with your arithmetic before you submit the other eight calls.
If your photo is a private file, host it at a public HTTPS URL first. Localhost and private-network URLs are rejected before submission.
A note on consistency across colorways
The product shape should not move between colorways. Send the same photo, the same aspect_ratio of auto, and a prompt that names only the color change, for example the swatch name and the finish. Anything else you add invites the model to redraw the object.
Two takes per colorway is a hedge against one bad render. At 2.1 1K the first-take-only run is 9 x $0.10 = $0.90 and the two-take run is 18 x $0.10 = $1.80. Decide from your own accept rate whether the second take pays for itself.
Name each file after the colorway and the job id when you store the output, so the billed amount can be traced back to the swatch.
Run the nine calls with distinct Idempotency-Key values if you send them in parallel, and reuse a key only when you retry the same payload after a timeout.
Sources
Related posts
More in Use cases
- 9-tile Instagram grid with 3 takes per tile: 27 images, from $0.67
A 9-tile grid with three takes per tile is 27 images: $2.70 on Nano Banana 2.1 at 1K or $0.67 on GPT Image 2.5 low on Sume. Arithmetic included.
- A 90-second avatar product video: two 45-second jobs, $0.22 to join
Sume avatar videos accept 4 to 60 seconds. For 90 seconds, run two 45-second jobs and join them: two audio detaches and one render add $0.22.
- 90-second explainer from fifteen 6-second scenes, plus the render
Fifteen 6 s scenes at 720p cost $11.25 on Wan 3.0 or Omni and $52.00 on Seedance 2.5; the 90 s timeline render adds $0.20 (2 billed minutes at $0.10).
- 90-second YouTube Short from three Wan 3.0 clips and a voice file
Build a 90-second vertical Short from three 30-second Wan 3.0 clips, one voice track and one Timeline render. Request shapes and the full cost at 480p to 1080p.
Written by Sume