Halloween product teaser: a 15-second vertical clip from one still

The quickest Halloween teaser is an image-to-video job: one clean product photo as the first frame, a prompt that describes the spooky motion, and seedance-2.5 at 9:16 for 15 seconds. Seedance 2.5 on Sume accepts 4 to 30 seconds at 480p, 720p or 1080p, takes a first and an optional last frame, and can generate an audio track. Halloween is Saturday 31 October, so there are four weeks to try variations.

Why a first frame

A text-only prompt invents your product. A first frame pins it: the label, colour and shape start exactly as photographed, and the model animates from there. Use a sharp, evenly lit photo with the product centred and spare background, because the opening frames are the part a viewer sees before they scroll. Send it in frame_images with frame_type: "first_frame". If you also send input_references, frame_images takes precedence and the request is treated as image-to-video.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: halloween-teaser-001" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "Fog rolls across a dark table, candle light flickers, the jar on the table glows softly orange, slow push-in, eerie low drone",
    "frame_images": [{
      "type": "image_url",
      "image_url": { "url": "https://example.com/jar.png" },
      "frame_type": "first_frame"
    }],
    "duration": 15,
    "resolution": "720p",
    "aspect_ratio": "9:16",
    "generate_audio": true
  }'

Read the price before you submit

Video prices differ by model and resolution, so we do not quote one here. Call GET /v1/videos/models and read pricing_skus for seedance-2.5. The listing also shows supported_durations, supported_resolutions and supported_aspect_ratios, so you can check a combination before you submit.

Alternatives by length

Finish the post

Poll GET /v1/videos/{id} until completed, then download from the content URL. For a branded end card, trim the best 12 seconds with POST /v1/video-trim ($0.02) and add a 3-second still in a Timeline 1.0 render ($0.10 per rounded-up minute). Add price or date text with caption cues, not inside the video prompt; models draw text unreliably.

Limits

A model will not guarantee that your label stays legible while the camera moves, so check the opening and closing frames before posting. Spooky atmosphere is easy; claims about the product are not the model's job, and you should keep them out of the prompt. Generated audio is a mood, not a mix: for a voiceover use TTS on a Timeline spine instead. Disclose AI-generated content wherever the platform asks you to.