App demo promo clip: generate the scene, composite the real screen

For an app promo, generate the scene around the phone with a video model, and put your real screen recording into it with Timeline compose. Compose stacks or overlays one still and one video, at $0.02 flat per shot, so the app UI is always the real one.

A generative model can draw a plausible app screen, but plausible is not true. A holiday promo that shows a button that does not exist is a support ticket.

What each Sume piece does

Sume's Timeline compose docs describe one still and one video on screen at the same time, returned as one MP4 shot. The operation field is stack or overlay. The output length comes from the video layer, the still stays on screen for the whole clip, and the ceiling is 300 seconds. Both URLs must already be this workspace's media.sume.com artifact or asset, so import your files first with POST /v1/media-imports. The Idempotency-Key is required.

The video step is a catalog model. Gemini Omni Flash 1.1 accepts up to three reference videos of up to 3 seconds each, and up to ten reference images, per the Video Router docs. Google's launch post (read 2026-10-05) says video references are up to 3 seconds as well. A short recording is therefore usable as a reference for the look of the screen, but treat it as guidance.

The plan

A four-step plan keeps the UI honest and the scene cheap.

Why a blank phone

The cost of the generated and composited shot is about $1.02, plus $0.10 for assembly of up to a minute of output. The recording is yours and free. For ten holiday variants of the same promo, you pay for ten scenes at $1.00, ten composes at $0.02 and ten assemblies at $0.10: $12.20.

Prompt the scene with a blank or dark phone screen, and keep the phone steady so the overlay lines up. If the phone moves, a static overlay will drift out of the frame. Either lock the camera in the prompt, or choose a layout in which the recording fills the frame next to the scene, which is the stack operation.

When a person helps

For a UGC-style app ad, a presenter holding the phone is more persuasive than a scene with no person. Sume's avatar product supports that look with an avatar and a scene prompt. The catalog Format named sume-mobile-app-ugc is also a candidate; read its description and io profile with GET /v1/formats/sume/sume-mobile-app-ugc before you call it.

Whichever route you use, keep the app store's own rules in mind for any clip that you reuse as a store preview. They set their own length and content rules, and those are on the store's help pages, not in this post.

What the finished promo looks like

Here is how the finished piece reads. Three seconds of the scene, a hand picking up a phone beside a holiday table. Four seconds of the phone filling the frame with your real recording of the key flow, such as ordering a gift in three taps. Two seconds of an end card with the offer. That is a nine-second promo, built from one generated shot, one composite and one assembly, and the only invented pixels are the table and the hand.

If the story needs a longer recording, trim it first. Sume has a video trim job at $0.02 flat that cuts [start, end) from one hosted clip into a new MP4, so you can cut the recording to the exact seconds you need before it goes into compose.

Keep the recording's aspect ratio in mind. A phone screen is tall. Place it in a 9:16 output with the stack operation if you want it large, or overlay it onto the scene if the scene should stay visible.