12-second AI video with first and last frame and sound under $2

Wan 3.0 at 480p ($0.75) or 720p ($1.50), H3 Max at 480p ($0.75) or 768p ($1.20), and H3 at 480p or 768p ($0.90) all produce a 12-second clip from a first and last frame with sound for under $2 on Sume.

The filter is narrower than it looks. A 12-second ask excludes Gemini Omni Flash 1.1, which stops at 10 seconds, and Kling 3.0 and Seedance 2 rows only stay under $2 at their cheapest settings.

The constraints

You need a duration of at least 12 seconds, a last_frame entry in frame_images, and audio. The Sume catalog says wan-3.0 is 2 to 30 seconds with image-to-video plus end frame and audio, minimax-h3 and minimax-h3-max are 5 to 15 seconds with the same modes and native stereo audio, and kling-3 is 4 to 15 seconds. Prices are list times 1.25, rounded up per job, per the Video Router docs.

Cost of 12 seconds, ranked

Under-$2 rows first, then the rows that miss. Seedance rows are per video token, at 16:9 and 24 frames a second.

Which to pick

For a preview, take the 480p rows: all three cost the same, so the tie-break is duration limits and features. For a final at a higher resolution, H3 Max 768p is the cheapest 768p row with a published refinement path to 1080p, and Wan 3.0 720p is a good choice when you may later extend past 15 seconds.

Remember that H3 and H3 Max use native 768p, not 720p, and that 1080p on H3 Max is a latent refinement of a 768p render. Treat the 12-second price as the reserve at submit time; the amount captured is the billable total for the job.

Request shape

{
  "model": "minimax-h3-max",
  "prompt": "The door opens, light floods the hallway",
  "duration": 12,
  "resolution": "768p",
  "frame_images": [
    {"type": "image_url", "image_url": {"url": "https://example.com/a.png"}, "frame_type": "first_frame"},
    {"type": "image_url", "image_url": {"url": "https://example.com/b.png"}, "frame_type": "last_frame"}
  ]
}