360p drafts before a 1080p final: when Omni drafting pays off
A 10 s Omni 360p draft costs $0.38 and a 1080p take $1.88. From the second attempt on, drafting first is cheaper. Break-even table on Sume, read 2026-10-05.

Draft at 360p first once you expect more than one attempt. A 10-second Gemini Omni Flash 1.1 clip costs $0.38 at 360p and $1.88 at 1080p on Sume, so a draft is about a fifth of a final take, and the draft path wins from the second attempt onward.
Google describes the 360p option as a draft mode that is "up to 60% faster and at a third of the cost" of standard 720p, meant for rapid prototyping and storyboard iteration (Google blog, read 2026-10-05). This post prices that idea on the Sume Video Router, where the model id is gemini-omni-flash-1.1.
The per-clip prices used here
Sume bills the provider list rate times 1.25 for each output second and rounds each clip up to the cent, so the per-second rates are $0.0375 at 360p, $0.125 at 720p, $0.1875 at 1080p and $0.375 at 4K. The ladder is documented in the Video Router page, with list rates of $0.03, $0.10, $0.15 and $0.30 per second.
| Resolution | List per second | Billed per second | 10 s clip |
|---|---|---|---|
| 360p | $0.03 | $0.0375 | $0.38 |
| 720p | $0.10 | $0.1250 | $1.25 |
| 1080p | $0.15 | $0.1875 | $1.88 |
| 4K | $0.30 | $0.3750 | $3.75 |
Break-even: direct finals versus draft-then-final
Suppose you need k attempts before a clip is acceptable. Going straight to 1080p costs k times $1.88. Drafting costs k drafts at 360p plus one 1080p final, so $0.38k + $1.88. The two lines cross at k = 1.25, which means one lucky first attempt favors going direct and every later attempt favors drafting.
The table below assumes the final render is the same request as the accepted draft with only the resolution changed. A new generation is a new job, so inspect the final before you ship it. The saving is real only when a draft predicts the final well enough to be worth approving.
| Attempts (k) | Direct 1080p | Drafts + 1 final | Saved by drafting |
|---|---|---|---|
| 1 | $1.88 | $2.26 | $-0.38 |
| 2 | $3.76 | $2.64 | $1.12 |
| 3 | $5.64 | $3.02 | $2.62 |
| 4 | $7.52 | $3.40 | $4.12 |
| 5 | $9.40 | $3.78 | $5.62 |
When not to draft
- You already know the prompt works, for example a re-run of a template with one swapped product photo. One direct final is cheaper.
- The check you care about is fine detail such as small on-screen text or skin texture, which a 360p frame cannot show.
- The clip is 3 seconds at 720p: it costs $0.38, the same as a 10-second 360p draft.
How to run it
Send the same body twice and change only resolution. Use a distinct Idempotency-Key per attempt so a network retry never double-bills, and keep the draft job id next to the final job id in your own log.
curl -X POST https://api.sume.com/v1/video-router/generate \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: spot-017-draft-1" \
-d '{
"model": "gemini-omni-flash-1.1",
"prompt": "A barista pours oat milk into a flat white, slow push-in",
"resolution": "360p",
"duration": 10,
"aspect_ratio": "9:16",
"mode": "async"
}'
# final: same body, "resolution": "1080p", Idempotency-Key: spot-017-finalWhat a draft should answer
A draft is only worth its price if it settles a question. Before you submit one, write the question down: does the subject stay on model, does the camera move do what the prompt said, does the action land inside the clip length, does the sound match the scene. A 360p clip answers all four. It does not answer whether faces are sharp or whether a label is legible, because 360p simply does not carry that detail.
That is why the break-even table is a floor, not a promise. If every draft answers one question and you need three answers, three drafts are the honest count and the cost is $1.14 for ten-second clips. Compare that with one blind 1080p try at $1.88 that answers the same three questions only if you are lucky.
Draft length matters as much as draft resolution
You do not have to draft at the final length. The Omni row accepts 3 to 10 seconds, and a 3-second clip at 360p is $0.12. If the question is whether a motion reads, a 3-second draft is a third of the price of a 10-second one. The trade is that you will not see the end of a long action, so draft at the full length when the ending matters, such as a reveal or a transition.
Some people draft at 360p and then jump straight to 4K for the final. At ten seconds that final is $3.75, so a three-draft path is $1.14 + $3.75 = $4.89. The same three attempts at 4K would be $11.25. The shape of the saving is the same as in the 1080p table, just larger in dollars.
Rerolls change the math
The table assumes the final is accepted on its first run. Finals can fail too. If one in three finals needs a re-run, add a third of the final price to every spot. At 1080p that is about $0.63 per ten-second spot. Even then, drafting at $0.38 a try keeps the total well under a path that rerolls at 1080p.
What this does not claim
Sume does not promise that a final matches its draft frame for frame; the docs describe the resolution tiers and prices, not draft fidelity. Treat the draft as a cheap vote on the prompt, then check the final.
Sources
Related posts
More in Pricing
- Omni 4K costs twice 1080p per second on Sume: is it worth it?
On Sume, Omni 4K is $0.375 a second against $0.1875 at 1080p: exactly double. A 10 s clip is $3.75 versus $1.88. When the extra pixels pay for themselves.
- Omni 9:16 and 16:9 360p drafts of one prompt: 75 cents for the pair
Gemini Omni 1.1 Flash takes only 16:9 and 9:16. Draft both at 360p for 10 seconds each, about $0.75 on Sume, before choosing which one to finish.
- One hour of avatar video: 60 one-minute jobs cost by tier
One hour of Sume avatar video at 60 one-minute jobs costs $662.40 standard, $882.00 plus or $1,980.00 max, before avatar creation and product options.
- One-minute AI avatar test: Griffin Turing call vs a 60 s Sume clip
Tavus tested Griffin-Lite in one-minute calls. Here is how a 60-second Sume avatar clip differs as a test, and what it costs at standard, plus and max.
Written by Sume