30 thumbnail concepts: draft at low, upscale 3 winners for $1.34
Draft 30 thumbnail concepts at GPT Image 2.5 low, upscale 3 winners at $0.20: $1.3425 total, against $6.675 at high or $6.7425 to upscale all.
Thirty thumbnail concepts drafted at GPT Image 2.5 low cost 30 x $0.02475 = $0.7425. Upscaling the 3 you pick adds 3 x $0.20 = $0.60, so the whole test costs $1.3425. Rendering all 30 at the high tier would be 30 x $0.2225 = $6.675, and upscaling every draft would be $6.7425.
The funnel
The idea is to spend the small amount on breadth and the large amount on a few finalists. A draft shows composition, face placement, text size and color contrast. These do not change when you upscale, so a concept that wins at draft size is still the winner at 3840 x 2160. Detail that only appears at high quality, such as skin texture, is what you give up in review.
| Plan | Calculation | Total |
|---|---|---|
| Draft 30 at low, upscale 3 | 30 x $0.02475 + 3 x $0.20 | $1.3425 |
| Draft 30 at low, upscale 5 | 30 x $0.02475 + 5 x $0.20 | $1.7425 |
| Draft 30 at low, upscale all | 30 x ($0.02475 + $0.20) | $6.7425 |
| Render all 30 at high | 30 x $0.2225 | $6.675 |
| Render all 30 at medium | 30 x $0.055625 | $1.66875 |
Reading the table
The first row is the cheapest way to get three large finals. The medium row is close at $1.66875 and gives you 30 medium images, so if you do not need 4K, you can stop there. The upscale-all row is slightly dearer than rendering at high, which means that you should never upscale the whole set. Upscale the finalists.
YouTube's help page recommends 3840 x 2160 for video thumbnails and keeps mobile uploads to 2 MB. Save the finalists as JPG if they come out larger than that.
Picking the three
Put the 30 drafts on one contact sheet and look at them at the size of a phone thumbnail. A good concept reads in under a second: one face or object, three or four words of text, strong contrast. Drop any draft that you have to squint at.
Pick winners before anyone sees the real video title, or you will choose thumbnails that fit the title and not the click. If you want to test them live, YouTube's own test tools are the next step. The upscale of three finalists is $0.60, so it is cheap to redo with a different set if the first test is flat. The media price sheet lists the same unit prices for larger volumes.
Scaling the funnel
The same shape works at other sizes. For 100 concepts, drafts are $2.475, and upscaling 5 finalists adds $1.00, so the test costs $3.475. For 12 concepts and 2 finalists, it is $0.297 + $0.40 = $0.697.
The more concepts you draft, the more the draft price matters and the less the finalist price does. Because the draft price is under three cents, breadth is nearly free. The finalists carry the cost, so keep their number small and chosen on purpose.
Sources
Related posts
More in Use cases
- 317 furniture photos: cutouts plus 40 hero upscales cost $15.13
Remove the background from 317 furniture photos and upscale the 40 hero shots: 317 x $0.0225 plus 40 x $0.20 is $15.13 on Sume. Line items and request body.
- 36 sneaker colorways from one photo: Nano Banana 2.1 edit, $3.60
36 colorway edits of one sneaker photo cost $3.60 on Nano Banana 2.1 at 1K on Sume: 9 calls of n=4. Table for 0.5K to 4K and a GPT Image 2.5 comparison.
- 37 event posters: 2 low drafts each, then Pro 2K finals, $8.77
Draft 37 posters twice on GPT Image 2.5 low ($1.83), then finish each on Nano Banana Pro 2K ($6.94): $8.77 on Sume. Arithmetic in the table.
- A 4-minute launch video as a TikTok ad: $0.40 to assemble
TikTok non-Spark ads allow 10 minutes; Spark Ads have no duration limit. A 4-minute launch video costs $0.40 to assemble on Sume Timeline 1.0.
Written by Sume