App preview clip from screenshots with Seedance 2.5 and a caption pass

A Seedance 2.5 clip can turn app screenshots into a lively 15-second promo, but it cannot be trusted to render your interface accurately. Send up to nine screenshots as reference_image_urls, describe a phone in a hand or on a desk, and burn the real copy on afterwards with an authored caption job. At 720p the 15 seconds cost about $8.67 on Sume's rate card.

What to expect from generated UI

A generative video model reinterprets the pixels it is given. Buttons may drift, labels may change and a chart may become a different chart. BytePlus's Seedance 2.5 page, read on 2026-10-04, says nothing about text or interface fidelity (BytePlus: Dreamina Seedance 2.5), so nothing here assumes the screens will survive intact.

Treat the generated clip as atmosphere: hands, desk, lighting and a glowing screen. Show the real, readable UI in a separate screen recording and join the two.

Request shape

Use a vertical aspect ratio for most app stores and ad placements. The Video Router docs list seedance-2.5 at 4 to 30 seconds and three resolutions. Keep the prompt focused on the scene around the phone.

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: app-promo-001" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "A person on a train scrolls through the app on a phone, soft window light, the screen glows, shallow depth of field.",
    "reference_image_urls": ["https://example.com/screen-home.png","https://example.com/screen-detail.png"],
    "resolution": "720p",
    "duration": 15,
    "aspect_ratio": "9:16",
    "mode": "async"
  }'

Add the words afterwards

A silent clip cannot be captioned from speech, so the video captions docs point to authored cues with text, start and end in seconds. Each standalone caption job is a fixed $0.20 for videos up to 60 seconds per those docs. The source must be a fetchable public HTTPS video URL.

Join with real footage

Assemble the generated clip and your screen recording with Timeline 1.0, which takes ordered video slots and returns one MP4. Set the output size to the placement you are delivering to, so a 9:16 clip is not cropped by a 16:9 default. Check the store's current video specifications yourself; this post does not state them.