Real estate walkthrough with no voiceover: burn room labels

To burn room names, price and address onto a silent real estate walkthrough, send POST /v1/video-captions with cues, each with text, start and end in seconds. Cues skip speech-to-text, so a clip with no voiceover works; a plain caption job on a silent clip fails with caption_no_speech. A job up to 60 seconds is $0.20.

The behavior below comes from Sume's video captions docs. Sume does not know the property, so you write every label; and it does not check listing rules, so confirm your portal's and local rules before you publish.

Why does a silent walkthrough fail a normal caption job?

Standalone captions run speech-to-text when you give no text, and speech-to-text needs audible speech. The docs say a silent clip fails as caption_no_speech with next_action: use_overlay_captions, not as a policy reject. That is a hint to switch to authored overlay copy.

A walkthrough that carries only music, or nothing, is a common way to make a listing clip, so cues are the route to use for it rather than an edge case.

The error is useful because it is typed. If your code sees caption_no_speech, it can branch to a cue path automatically instead of failing the batch, and a listing team that captions fifty walkthroughs at once will meet it often.

What does a cue-based caption job look like?

A cue is phrase-level overlay text with a start and an end. cues and segments are the same idea; script_text, words, cues and segments are mutually exclusive, so pick one. The body must be a public HTTPS video_url. Add an Idempotency-Key so a retry does not bill twice.

Pick a style that stays readable on bright interiors. black-outline is the safe default for Hangul; for Latin text, name slam or punch, or omit style and Latin gets slam. design.placement.anchor_ratio moves the line centre as a fraction of frame height, which helps keep a label off the floor of a vertical clip.

Keep each cue to a few words. The docs list phrasing overrides such as max_words and max_chars, but cues are already phrase-level overlay cards, so the simplest fix for a long label is a shorter label.

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: listing-48-oak-cues-001" \
  -d '{
    "video_url": "https://example.com/listings/48-oak/walkthrough.mp4",
    "style": "slam",
    "cues": [
      { "text": "48 Oak Street", "start": 0.0, "end": 3.0 },
      { "text": "Open kitchen", "start": 4.0, "end": 8.0 },
      { "text": "Primary suite", "start": 12.0, "end": 16.0 },
      { "text": "Offered at $650,000", "start": 26.0, "end": 30.0 }
    ]
  }'

How should the labels be timed?

Watch the clip and note the second each room first appears. Start a label half a second after the cut so the viewer sees the room first, hold it for 3 to 4 seconds, and keep price and address to the start and end, not the whole clip. Fewer, shorter words read better on a phone.

If the clip is longer than 60 seconds, cut it first with video trim or caption it in parts, because the $0.20 fixed estimate covers videos up to 60 seconds and live pricing should be confirmed in GET /v1/catalog.

What does this not cover?

Burned-in text cannot be edited later: change a price and you re-run the job from your clean source video, so keep that source file. Sume does not publish to a listing portal, and portals have their own video rules on orientation, length and branding that this page has not verified, so read the portal's current guidelines first.

If the property photos were AI-edited, disclosure rules can apply; see California AB 723 and AI-edited listing photos. For turning stills into a narrated tour instead, see photos to a narrated tour with TTS and Timeline.

Check numbers twice. A price or square-footage typo in a burned-in label is a listing error you cannot patch without a new render, so read every cue aloud against the listing sheet before you submit the job.