Generate at 1K and upscale, or generate 4K? Sume price check
Nano Banana 2.1 at 1K plus a Sume image upscale is $0.30, against $0.20 for native 4K. GPT Image 2.5 low plus upscale is $0.22 against $0.22 for high 4K.
Generating at 1K and then upscaling does not save money on Sume's current prices. Nano Banana 2.1 at 1K ($0.10) plus an image upscale ($0.20) is $0.30 per finished image, while native 4K from the same model is $0.20. For GPT Image 2.5, low quality at 1K ($0.02475) plus the upscale is $0.22475, against $0.2225 for high quality at 4K.
The arithmetic
Sume charges a flat $0.20 per image for its upscale job, and its catalog reserves roughly 16 megapixels of generative output for it. The image model rows are list prices from the same catalog.
| Path | Step prices | Total |
|---|---|---|
| Nano Banana 2.1, native 4K | $0.20 | $0.20 |
| Nano Banana 2.1, 1K then upscale | $0.10 + $0.20 | $0.30 |
| GPT Image 2.5, high 4K | $0.2225 | $0.2225 |
| GPT Image 2.5, low 1K then upscale | $0.02475 + $0.20 | $0.22475 |
Why the second pair is a near tie
The GPT Image 2.5 pair differs by $0.00225, which is within noise once a single reroll costs more than that. The upscale route can still make sense there for a different reason: a low-quality 1K draft costs $0.02475, so you can generate several drafts, pick one, and pay for the upscale only on the winner. Ten drafts and one upscale cost $0.4475, against ten high 4K images at $2.225.
When each path wins
Native 4K is the better default when every image ships. The draft-then-upscale path wins when you throw most drafts away. It also lets a person choose the composition on cheap 1K files before any money goes into resolution.
One caution. This page compares prices, not pixels. An upscaled image and a natively generated 4K image are different files, so open both before you swap pipelines. The Image API page lists the model parameters, and the public catalog lists the upscale job's price.
- All shipped images are final: generate native 4K.
- Many drafts, few winners: draft cheap, upscale the winner.
- Always compare an upscaled file and a native one by eye first.
Reproduce the numbers
Every Sume price on this page comes from the public catalog at https://api.sume.com/v1/catalog, which needs no API key. Fetch it, find the row for the job or model named in the table, and multiply by your own seconds, characters or image count. Sume bills the list price shown in that row, so the catalog figure is the one that applies to your balance.
All figures on this page are Sume's own, as of 2026-10-09. Prices and limits can change, so re-read the catalog and the linked docs before a large run, and keep the arithmetic in a spreadsheet so you can update it in a minute.
Sources
Related posts
More in Comparisons
- GPT Image 2.5 on Sume: low 1K $0.02475 vs high 4K $0.2225
GPT Image 2.5 spans 9x in price on Sume: $0.02475 for low at 1K, $0.055625 medium at 2K, $0.2225 high at 4K. Omit quality and the default is high.
- Griffin-Lite's 0.43 s latency: what a recorded avatar clip gives up
Tavus reports Griffin-Lite video latency of 0.43 s on average (0.27-0.59 s) in a research preview. A recorded clip on Sume is a job, so choose by use case.
- HappyHorse 1.0 at $0.15 a second: four Sume rows cost less at 720p
Runway Dev lists HappyHorse 1.0 at 15 credits per second at 720p and 30 at 1080p. Sume does not list it; four Sume rows are cheaper at 720p, three at 1080p.
- HappyHorse 1.1 lists 9 aspect ratios; Omni on Sume lists 2
HappyHorse 1.1 reference-to-video offers nine ratios, from 9:21 to 21:9. Gemini Omni Flash offers 16:9 and 9:16. How to check a Sume model's ratios first.
Written by Sume