AI video generator using a reference video: Seedance 2.5 setup, cost

To generate video from a reference video, send the clip as a video_url entry in input_references on seedance-2.5 and say in the prompt what to take from it, such as the camera move or the pacing. A reference video also changes the price: Sume's estimate adds about 15 seconds of input and applies a 0.6 multiplier to the total.

The vendor limits for reference videos come from BytePlus's Seedance 2.5 tutorial.

How do I send a reference video?

Use a public HTTPS URL. Sume's seedance-2.5 catalog lists video_url as a supported reference type.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "A sneaker on a plinth, camera move copied from @video1",
    "duration": 10,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "input_references": [
      {"type": "video_url", "video_url": {"url": "https://example.com/move.mp4"}}
    ]
  }'

What are the reference video limits?

BytePlus lists up to 10 videos, each 2 to 30 seconds and up to 200 MB at 24 to 60 fps, with a total of up to 30 seconds. Its prompt guide suggests 5 to 10 seconds.

How does a reference video change the cost?

Compare the same request with and without a reference, for seedance-2.5 at 720p and 16:9.

Why is the 30 s reference row lower?

The estimate treats the reference as about 15 extra seconds of input, then multiplies the token total by 0.6. For a 30-second output that is (30 + 15) × 0.6 = 27 billed seconds, so it can land below the no-reference figure. Treat it as an estimate and check the job's actual charge.

Does the reference edit my video in place?

No. On Sume the clip is a reference for a new generation. For a prompt-based edit of an existing video, see edit a video with a prompt.