Handheld selfie UGC prompt for Seedance 2.5: a template and cost

Write a handheld selfie UGC prompt as five short lines: who is holding the phone, what the lens sees, what the hand does, what is said or heard, and how the clip ends. Send it to seedance-2.5 with aspect_ratio: "9:16" and a 10-15 second duration.

At 720p a 10-second vertical clip is $5.78 on Sume and a 15-second clip is $8.67. The template below keeps a 480p draft at $2.69 for ten seconds, so a bad prompt costs little.

The five-line template

Each line answers one question the model would otherwise guess. Keep each to one sentence.

• Holder: "A woman in her thirties holds the phone at arm's length, slight handheld sway."
• Frame: "Close selfie framing, face and shoulders, a kitchen counter behind her, daylight from the left."
• Hand action: "With her free hand she lifts a small blue jar into frame and tilts the label toward the lens."
• Sound: "She says: this one actually stays on. Quiet room tone, no music."
• Ending: "She smiles, lowers the jar, and the clip ends on the label."

Why the lines are in this order

Handheld UGC fails when the model has to invent the camera. Naming the holder and the sway first fixes the camera before the subject. The frame line then fixes distance and light, which are what make a clip read as phone footage instead of studio footage.

Put speech in quotes and keep it to a single short sentence. Seedance 2.5 is described by ByteDance Seed as an audio-video joint generation model, and Sume lists audio as available on the seedance-2.5 row, so spoken and ambient sound belong in the prompt rather than in a later step. Check the audio result before approval, since a lip that does not match a line is a re-render, not a trim.

Settings and price

Vertical output uses aspect_ratio plus resolution; Sume does not accept a pixel size for these models. The price depends on resolution and length, not on the aspect ratio, so a 9:16 clip costs what a 16:9 clip of the same length costs.

Submit the request

The Idempotency-Key header makes a retry safe: a replay returns the original job instead of charging a second one.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: selfie-draft-001" \
  -d '{"model":"seedance-2.5","prompt":"Handheld selfie video, a woman in her thirties holds the phone at arm's length with slight sway. Close framing, kitchen counter behind, daylight from the left. She lifts a small blue jar into frame and tilts the label to the lens. She says: this one actually stays on. Quiet room tone. She smiles and the clip ends on the label.","duration":10,"resolution":"480p","aspect_ratio":"9:16"}'

What to change on the second take

Change one line per take. If the hand action is wrong, rewrite only that line and keep the other four. Because Sume's video models do not accept a seed, a re-render is a new take and the other four lines may look different anyway, so compare takes by the line you changed, not by pixel match.

If you have a real photo of the creator or the product, send it as an image reference instead of describing it. Sume's request validation accepts up to 9 reference images for Seedance 2.5, and the docs state that a model takes a reference type only if its supported_input_references lists it. Send only images you have the right to use.

Draft cheap, finish once

Run the template at 480p first. Ten 10-second drafts are $26.90; one 15-second 720p keeper is $8.67. Move to 1080p only for the final export, and only if the placement needs it. For polling the job, see the video generation docs: the status moves from pending to in_progress to completed, and the video downloads from the content URL.