News summary video with AI: Seedance 2.5 b-roll, captions, one MP4

To make a news summary video with Seedance 2.5, write the script first, generate one 4 to 30 second illustration clip per story beat, record or generate the voiceover, burn your own caption lines, and join everything in a Timeline render. BytePlus names "News summary and storytelling" as a use case on its Seedance 2.5 page. The model supplies moving illustration; it does not supply facts, and a clip made from a prompt is not footage of an event.

What does the vendor page say?

The BytePlus Seedance 2.5 page, read 2026-10-03, lists four use cases: film and video production, news summary and storytelling, product marketing videos, and previsualization and pitch videos. It describes 30-second generation in one pass with room for "emotional beats, product walkthroughs, or narrative progression", and support for images, video and audio references.

That tells you what the vendor aims at. It does not make generated clips a source of news. Keep real photos and verified footage for the facts and use generated clips as background visuals.

How do you structure the recap?

Write the script as numbered beats of 5 to 15 seconds, one fact per beat. For each beat decide what is on screen: a real image you already own, or a generated clip. For generated clips, describe a place or mood, not a named person or an event you could be accused of staging.

Generate clips at 480p first. The Video Router doc says limits are per model, so read capabilities from GET /v1/video-router/models rather than assuming; for seedance-2.5 the minimum is 4 seconds, so a 3-second beat needs a longer clip that you trim, or a slot with duration set on the Timeline.

How do you add headlines without speech-to-text?

Video captions burns text onto a public HTTPS video URL. If you pass cues (or segments) with text, start and end, it burns that authored copy and skips speech-to-text, which suits headline cards and source lines. Without cues, a clip with no speech fails with caption_no_speech.

Burn the captions on the finished cut, so the lines line up with the voiceover; if your Timeline output URL is not reachable by the caption job, check the doc's input rules before you build the step in.

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: recap-captions-001" \
  -d '{
    "video_url": "https://example.com/recap.mp4",
    "style": "punch",
    "cues": [
      {"text": "Rates held steady", "start": 0.5, "end": 3.0},
      {"text": "What changes for renters", "start": 3.2, "end": 7.0}
    ]
  }'

What does the Timeline step look like?

A recap of 45 seconds needs an audio.duration_seconds of 45 and a list of slots whose start values increase from 0. Declared starts are authoritative, and the doc says coverage may trail the spine by at most 0.5 seconds, so plan clip lengths to cover each beat. Use transition on every slot after the first only where it helps: a fade of 0.25 seconds reads as a cut between stories, while a slideup can mark a new section.

Timeline's output is 1080 by 1920 by default. For a landscape recap, set output.width and output.height to even values between 256 and 2160. Run the unbilled plan call first and read segment_count and estimated_cost_usd_micros; render only when they match what you expect. Timeline is billed at $0.10 per rounded-up output minute, per its doc; confirm the live rate in the catalog.

What should a newsroom check?

Label generated visuals so viewers know they are illustrations. Check each clip frame by frame for invented text, logos or faces. Keep the script, sources and the request body for each clip with the final file. Seedance has no seed on Sume, so you cannot regenerate the identical clip later; the stored output is the record.