Veo 3.1 1080p needs 8 seconds; Omni on Sume does 3 to 10 at 1080p

Veo 3.1 only produces 1080p and 4K when the clip is exactly 8 seconds, but Gemini Omni on Sume renders 1080p at any whole length from 3 to 10 seconds. A 5-second 1080p Omni clip lists at $0.75 and bills $0.94 after the 1.25 multiplier and rounding up.

The Veo rule

Google's Veo guide says durations are 4, 6 or 8 seconds, and that the duration must be 8 seconds for 1080p, 4K, reference images or extensions. Extensions are limited to 720p. So a 4-second social cut at 1080p has to be trimmed from an 8-second render, and you pay for all 8 seconds.

Cost of short 1080p clips

Veo prices come from Google's pricing page: Standard $0.40 per second at 720p and 1080p, Fast $0.12 at 1080p, Lite $0.08 at 1080p. Omni rows use Sume's list of $0.15 per second at 1080p times 1.25, rounded up to cents per job.

Compute it

This prints the Omni column of the table from the documented rule, so you can change the length or the resolution rate.

import math

LIST_1080P = 0.15  # Sume catalog list per second for gemini-omni-flash-1.1

for seconds in (3, 5, 8, 10):
    billed = math.ceil(round(seconds * LIST_1080P * 1.25 * 100, 6)) / 100
    print(f"{seconds:>2} s  list ${seconds * LIST_1080P:.2f}  billed ${billed:.2f}")

Picking a length

On Veo, a shorter clip at 1080p means paying for 8 seconds and cutting. On Omni through Sume, you pay for the length you ask for, which is the main reason the 3-second row of the table above costs $0.57.

• Use 3 to 5 seconds for a hook or a loop.
• Use 8 seconds for a standard product beat.
• Use 10 seconds when a gesture needs room to finish.
• Remember that edit mode takes its length from the source clip, so the length rule applies to generation only.

Trade-offs

Veo 3.1 Fast at 8 seconds is cheaper per clip on Google's page ($0.96 against $1.50), but it is a preview id that shuts down on October 22, 2026. If the exact length drives your edit, Omni's flexible duration saves the trim and the wasted seconds. Always send duration as an integer in the 3 to 10 range, and read the supported_durations of the row in the Video Router catalog before a batch.