SaaS app UGC video on Seedance 2.5: screen reference and cost

A generated clip cannot show your real interface. Use Seedance 2.5 for the person and the scene, and add the real screen in an editor. A 15-second 9:16 UGC clip costs $8.67 at 720p on Sume.

Text on screens is a weak point for video models. A screenshot sent as a reference guides the look; it does not guarantee readable copy or the exact layout. Treat any on-screen text in the output as decorative.

What to ask the model for

Ask for a person at a desk with a laptop angled so the screen is not readable, reacting to the result. Ask for the phone held toward the camera with the screen glowing. Then cut the real recording over those moments. The generated part supplies the human reaction and the setting, which a screen recording does not have.

Prompt blocks

• Scene: "A tidy home office, soft window light, a person in a plain navy hoodie at a laptop."
• Action: "They click, lean closer, then smile and nod at the screen."
• Framing: "Over-the-shoulder phone-style shot, the laptop screen is angled away so it is not legible."
• Sound: "Quiet room tone, a soft keyboard click."
• Avoid: any request to show specific UI text, charts with numbers, or a logo.

Price at the lengths people use

Most app clips are 10-20 seconds.

Request

No reference is needed for this version because the screen is not shown.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: saas-ugc-001" \
  -d '{"model":"seedance-2.5","prompt":"A tidy home office in soft window light. A person in a plain navy hoodie clicks at a laptop, leans closer, then smiles and nods. Over-the-shoulder phone-style shot, the screen angled away and not legible. Quiet room tone, a soft keyboard click.","duration":15,"resolution":"720p","aspect_ratio":"9:16"}'

Cutting the real screen in

Generate the clip with a laptop angled away. In your editor, track a rectangle over where the screen would sit, or simply cut away to the recording at the moment the person reacts and cut back for the smile. The cut-away is easier and looks natural, because UGC videos switch between the creator and the screen all the time.

Match the sound. Fade the room tone under the screen recording so the cut does not jump. Add captions in the editor, not in the prompt; a caption written by a video model will be misspelled often enough to matter, and you will want the product name exact.

Choosing a length

Ten seconds is enough for one reaction, 15 for a reaction and a second beat, and 20 only if the person has to do something longer. Longer clips give more room for the person to drift, so start at 10 seconds ($2.69 at 480p) and extend only if the story needs it.

If you want the real UI in the clip

Two routes exist. The first is the editor route above: composite your screen recording over the generated laptop. The second is to send your screenshot as an image reference and accept an approximation. Sume's request validation accepts up to 9 images, 3 videos and 3 audio files on Seedance 2.5, but the output is a rendering of an app, not your app.

Do not publish a generated screen as a real product shot. It can mislead viewers about features, and that is a trust problem for a software ad. If the clip's claim is about what the product does, show the product recording for that part.

For an editing approach on an existing take, Sume documents gemini-omni-flash-1.1 accepting a video_url to edit a clip, which is a separate model with its own price and limits. Test it on a short sample before it enters a workflow; do not assume it can replace a screen.