Pop-up shop teasers: one still, three 5-second videos, dated cues

Plan a three-day pop-up shop teaser series on Sume: one Nano Banana 2.1 still, three Wan 3.0 jobs, three caption cue jobs. Chain and limits, read 2026-10-10.

4 min readSume
All posts

A pop-up shop opening can be teased for three days for about $2.58 on Sume: one Nano Banana 2.1 still ($0.10), three 5-second Wan 3.0 clips at 720p ($1.875), and three overlay-caption jobs ($0.60) that print Thursday, Friday and Saturday text on the clips. The dates are text, not pixels the model drew.

Pop-ups are short and local, so the post must name the day and the street; image and video models still misspell both. The workable pattern is a generated mood with authored overlay captions on top. Prices are Sume billable amounts from the pricing code (read 2026-10-10).

What is the endpoint chain?

Make the still once, then fork it three ways.

  • POST /v1/images (google/nano-banana-2.1, 9:16, 1K): the shop interior as a mood, no signage text.
  • POST /v1/videos three times with wan-3.0, resolution: 720p, duration: 5, frame_images with the still as first_frame and a different motion prompt each day (a door opening, a rail of coats, a counter pan). Wan accepts 2 to 30 seconds; 5 is plenty for a teaser.
  • POST /v1/video-captions three times with cues such as 'Opens Saturday' on day three. Send a distinct Idempotency-Key for each of the six writes.

What does it cost?

Wan 3.0 bills per second by resolution: $0.0625 at 480p, $0.125 at 720p, $0.25 at 1080p. At 720p a 5-second teaser is $0.625.

Pop-up teaser series, Sume billable prices (read 2026-10-10)
StepBasisCost
Mood still1 x 1K$0.10
Three teasers3 x 5 s x $0.125$1.875
Date cues3 caption jobs x $0.20$0.60
Total$2.575
Same at 480p3 x 5 s x $0.0625 + the rest$1.6375

What are the limits?

Caption cues need text, start and end in seconds and cannot be mixed with script_text; a silent clip is exactly the case they are for. A clip that has audio but no speech fails speech-to-captions with caption_no_speech, so do not ask for auto captions here.

Wan will not give the same room twice from text alone; the first frame is what keeps the three days looking like one place. Wan also accepts reference images, so if the rack matters, reference a photo of it. Check the live ranges with GET /v1/videos/models and read the Video Generation page for frame_images.

Sume does not take a seed on any v1 video model, so a replay of the same prompt will not be byte-identical; keep the winning file instead of regenerating.

Should the dates be live text or burned in?

Burned in is the only choice for a downloaded MP4, and the caption page is where the styles and design overrides live. If a date changes after rendering, recaption the clean clip (a $0.20 job) instead of regenerating it. The same idea runs a 12-item bulk queue when the series is longer.

Scheduling the three days

Generate all three teasers together on the first day; the dates on each one are cues, so a delay in opening only means recaptioning the three clean clips. That is three $0.20 jobs, not three new video jobs.

If you post the teasers through a social tool, upload the final MP4s from their media.sume.com URLs, which are the durable copies.

  • Day one: a rail and a door, cue 'Opening Saturday'.
  • Day two: the counter, cue 'Tomorrow, 10 to 6'.
  • Day three: a wide shot, cue 'Today' with the street name.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume