Nano Banana 2.1 0.5K is a 512 tier; YouTube thumbnails need 640 wide
The Image API names its smallest tier 512, below YouTube's 640-pixel minimum thumbnail width. 91 test thumbnails cost $6.825 at 0.5K, $9.10 at 1K, $13.65 at 2K.
The cheapest Nano Banana 2.1 row on Sume is the 0.5K tier at $0.075 per image, and the Image API docs name the matching normalized resolution tier 512. YouTube Help (read 2026-10-09) states a minimum width of 640 pixels for video thumbnails. If the 512 tier means a 512-pixel edge, a 0.5K image is below that minimum, so use it for private drafts and move to 1K ($0.10) for anything you upload.
A creator who tests 7 thumbnail variants on each of 13 videos makes 91 images. That costs 91 x $0.075 = $6.825 at 0.5K, 91 x $0.10 = $9.10 at 1K and 91 x $0.15 = $13.65 at 2K.
What YouTube states
YouTube recommends a thumbnail of 3840 x 2160 for videos, a minimum width of 640 pixels, JPG or PNG, and 16:9. It also states a 2 MB mobile upload cap and a 50 MB desktop cap. None of this says a small file is rejected, only that 640 is the stated floor and bigger is better. The embedded player also uses the thumbnail as its preview image, and YouTube says the image should be as large as possible.
| Tier | Price per image | Arithmetic | Total |
|---|---|---|---|
| 0.5K (512 tier) | $0.075 | 91 x 0.075 | $6.825 |
| 1K | $0.10 | 91 x 0.10 | $9.10 |
| 2K | $0.15 | 91 x 0.15 | $13.65 |
| 4K | $0.20 | 91 x 0.20 | $18.20 |
A two-stage plan
The saving from drafting small is real but modest: 91 drafts at 0.5K instead of 1K saves 91 x $0.025 = $2.275. The better plan is to draft at 0.5K, pick the winner for each video, then regenerate only 13 finals at 2K or 4K. That is 91 x $0.075 + 13 x $0.15 = $6.825 + $1.95 = $8.775, versus $13.65 if every variant is made at 2K.
Draft and final images from different tiers will not match exactly even with the same prompt, so treat the draft as a layout check and write the final prompt from what the draft showed. Look at the returned image dimensions before you upload; Sume's docs say the tier is a normalized setting, and the file you receive is what YouTube will see.
Format and ratio
Ask for aspect_ratio 16:9 and output_format jpeg or png. The Image API docs list png, jpeg and webp as output options for models that accept the field, and YouTube accepts only JPG and PNG, so avoid webp. Keep the headline text large and the face centered so the thumbnail reads at small sizes.
Estimate before the first run
The plan above assumes every variant is kept. If you cull weak prompts after the first 30 drafts, the volume falls. Thirty 0.5K drafts cost 30 x $0.075 = $2.25, and if only 10 survive into full runs of the remaining 61 variants, you still save money against drafting everything at 2K.
Use the response field usage.cost to confirm the actual charge per tier on your first five images, and stop if it differs from the price list. Sume's Image API docs say cost is the USD amount billed to your wallet, and failed generations are not billed, so a failed draft costs nothing.
Sources
Related posts
More in Use cases
- 120 Nano Banana 2.1 2K hero images: $6.05 at Google, $18.00 on Sume
Google lists Nano Banana 2.1 at $0.0504 per 2K image; Sume lists $0.15. For 120 listing hero images that is $6.05 versus $18.00, plus what the difference buys.
- Nano Banana 2.1 product photo 1:1: 1024, 2048, 4096 and cost per tier
Square product shots with Nano Banana 2.1: 1024, 2048 and 4096 px sizes, what each costs on Sume, and what you cannot do, such as transparent output.
- Nano Banana 2.1 still, then Omni 1080p for 6 s: $1.225 together
A Nano Banana 2.1 still at $0.10 and a 6-second Omni Flash 1080p animation at $1.125 total $1.225 on Sume. The two-call Python script and where the cost sits.
- Narrate a Claude Motion explainer with Sume TTS and Timeline: 12 cents
A 450-character voiceover costs $0.021 on Sume TTS and a one-minute Timeline render $0.10, so narrating a Claude Motion MP4 comes to about $0.12.
Written by Sume