Which image model for thumbnails? Six Sume picks from 3 to 10 cents

Six Sume image models that list 16:9 and 4:5 for thumbnails and covers, from Qwen Image at 3 cents to Nano Banana 2.1 at 10, with the cost of 12 options.

5 min readSume
All posts

For thumbnails on Sume, start with six models that list 16:9, 9:16 and 4:5: Qwen Image (3 cents), Flux 2 Pro (4), Seedream 5.0 Lite (5), GPT Image 2.5 at high (7), Ideogram 4.5 at medium (8) and Nano Banana 2.1 (10). A set of 12 options per video costs from $0.36 to $1.20. Sume does not publish a quality ranking, so test two or three on your own style.

Cost of a 12-option set

Sume bills the catalog list price multiplied by 1.25 and rounded up to the next cent. Twelve options is three calls with n set to 4.

Billed cost for 12 thumbnail options and for 100 videos, as of 2026-10-08
ModelBilled per image12 options100 videos x 12 optionsTakes references
Qwen Image3 cents$0.36$36yes
Flux 2 Pro4 cents$0.48$48yes
Seedream 5.0 Lite5 cents$0.60$60yes
ChatGPT Image 2.57 cents$0.84$84yes
Ideogram 4.58 cents$0.96$96yes
Nano Banana 2.110 cents$1.20$120yes

Pick by the job

Thumbnails with a face from your own photo need references, so choose from the edit-capable rows. GPT Image 2.5 takes up to 16 references, and Ideogram 4.5 takes 5 total. For lettering you will be changing often, the Ideogram 4.5 docs say an edit without aspect_ratio keeps the shape of the source image, so a headline revision does not change the frame.

For pure volume tests, draft at Qwen Image or Flux 2 Pro, pick the composition, then re-render the winner at higher cost. The ratio is set per request, so 16:9 for the video and 4:5 or 9:16 for a social cutdown are one parameter apart.

  • Cheapest 16:9: Qwen Image at 3 cents.
  • Most references: GPT Image 2.5 at 16.
  • Most ratios for odd placements: Ideogram 4.5 lists 15, including 16:10 and 3:1.

Check before you publish

Look at the result at about 168x94 pixels, the size of a small suggested-video tile. Text that survives there will survive anywhere. Rejected ratios return 400, so read supported_parameters first.

A thumbnail workflow

Generate 4 options per call at n=4 with a 16:9 ratio. Choose the best, then re-run it as an edit with your face photo or brand art as a reference to fix identity. Keep headline text short, since a thumbnail is read at a small size, and add exact wording after generation if a misspelled word would be costly. Budget the whole loop: three n=4 calls on Seedream 5.0 Lite cost 12 x 5 = 60 cents, and one polish edit at 7 cents on GPT Image 2.5 brings the video to 67 cents.

Sources

Related posts

More in Models

All Models posts

Written by Sume