3-minute Reel from API clips: calls and cost by Sume video model

Instagram Reels now run up to 3 minutes, as reported by NapoleonCat (read 2026-10-07). A 180-second vertical video is longer than any single clip a Sume model returns, so the question is how many API jobs it takes and what they cost. The answer depends on each model's longest clip: 10 seconds for Gemini Omni Flash 1.1, 15 for Kling 3 and MiniMax H3, 30 for Wan 3.0 and Seedance 2.5.

The table prices a full 180 seconds of 9:16 footage with no retakes. It uses the catalog list rate times 1.25, and rounds each job up to the cent, the way Sume bills.

180 seconds of vertical video, one pass, no retakes

What the number of jobs changes

Fewer jobs means fewer seams. Each clip starts from a new generation, so a face, a wall colour or a product label can drift at every cut. Six 30-second jobs give you five seams; eighteen 10-second jobs give you seventeen. If a character has to hold across the whole Reel, prefer the models with the longest clip and feed the previous last frame to the next job as first_frame, or pass the same reference images to every job where the model accepts them.

Audio adds a second constraint. Omni Flash and the MiniMax H3 models always generate native audio, and Wan 3.0, Seedance 2.5 and Kling 3 let you choose with generate_audio. When you join clips, music and room tone restart at every cut unless you replace the audio track with one bed. For that case, send generate_audio: false where the model allows it and add a single music track afterwards.

Submit the six Wan 3.0 jobs

Every job is an independent POST /v1/videos. The Idempotency-Key header makes a retry safe: a replay returns the original job instead of billing a second one.

Check supported_durations in GET /v1/videos/models before you pick the length; Wan 3.0 accepts 2 to 30 seconds.

for i in 1 2 3 4 5 6; do
  curl -s -X POST https://api.sume.com/v1/videos \
    -H "Authorization: Bearer $SUME_API_KEY" \
    -H "Content-Type: application/json" \
    -H "Idempotency-Key: reel-0042-shot-$i" \
    -d '{"model":"wan-3.0","prompt":"Shot '$i' of a street-food vendor, handheld, warm light","duration":30,"resolution":"720p","aspect_ratio":"9:16"}'
  echo
done

Which route to pick

For the lowest bill, Wan 3.0 at 720p or MiniMax H3 at 768p land at about $22.50 and $13.56 for the whole Reel. For the most sound-bearing footage per call, Kling 3 with audio on costs $37.80. Seedance 2.5 costs the most here because its price scales with pixels, but it is also the model that takes image, video and audio references in one request. See the Seedance 2.5 per-second price before you choose it for a long edit.