Podcast teaser clip: cover art plus audio reference on Seedance 2.5

A podcast teaser on Seedance 2.5 needs a reference image, such as cover art or a host illustration, plus the audio excerpt in reference_audio_urls; Sume rejects audio on its own. Expect an atmospheric animated clip around your sound, not a guaranteed talking head. For speech that must match a face, Sume's Avatar Video is the purpose-built route. A 15-second 9:16 teaser is about $8.67 at 720p.

What Seedance does and does not promise

ByteDance describes Seedance 2.0 as a unified audio-video joint generation model that takes text, image, audio and video inputs (ByteDance Seed: Seedance 2.0). That page gives no lip-sync accuracy figure, so this post makes no lip-sync claim. If the teaser must show a speaker whose mouth matches the words, test it first or use a different tool.

The two routes on Sume

Pick by what the viewer must see. Both are documented, and the choice is about the face.

Build the teaser clip

Cut the excerpt first, using video trim if you start from a longer file, because a short, clean excerpt is easier to reason about than a full episode. Then send the cover art and the excerpt together. Sume allows up to 3 audio references per Seedance request.

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: podcast-teaser-001" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "Soft animated waveform lights pulse around the illustrated hosts, slow push-in, warm studio colors.",
    "reference_image_urls": ["https://example.com/cover-art.png"],
    "reference_audio_urls": ["https://example.com/excerpt.mp3"],
    "resolution": "720p",
    "duration": 15,
    "aspect_ratio": "9:16",
    "mode": "async"
  }'

Cost and next step

The 15-second 720p clip is about $8.67 and the 30-second version about $17.33, computed from Sume's token-based pricing at list x 1.25. Add captions with the episode title and a listen link as authored cues, a flat $0.20 job. Listen to the result before posting; generated audio treatment is not something to assume.