20 AI b-roll clips: cost by model and resolution on Sume

Twenty six-second b-roll clips (two minutes of footage) cost between $1.50 and $170.59 on Sume depending on model, from Grok Imagine Video 1.5 (image-to-video only) at the low end to Seedance 2.5 at 1080p at the top; Seedance 2.5 at 720p is $69.34. A mainstream choice, Wan 3.0 or Gemini Omni Flash at 720p, costs $15.00 for the pack.

These are billable amounts: provider list times 1.25, charged per output second. The clip length of six seconds is our planning choice and fits every model in the table: Omni takes 3 to 10 seconds, MiniMax H3 5 to 15, Grok Imagine Video 4 to 15 and Seedance 4 to 30. Rates were read from the pricing package and the Video Router docs on 2026-10-03.

Price for the pack, by model

The per-second rate is the number to compare. The Seedance rows come from token pricing: width times height times seconds times 24 frames per second, divided by 1,024, at the model's rate per 1,000 video tokens. Gemini Omni Flash and Wan are flat per second by resolution. Native audio is always included with Omni.

The Grok Imagine Video 1.5 row is image-to-video only: it needs a start image (image_url or first_frame_url), takes no reference media, and lists 480p and 720p at a flat per-second rate. It suits animating stills as b-roll, not text-to-video.

Takes per keeper

No one keeps every b-roll clip. If two takes are needed per keeper, a 20-keeper pack costs twice the sticker price; at three takes, three times. The table prices the models where the difference matters most.

For the mainstream models the pack is $30.00 at two takes and $45.00 at three. For Seedance 2.5 at 720p, two takes is $138.67.

Which model for which job

A b-roll shot behind a voiceover can tolerate a cheaper model; a hero shot cannot. A practical split is to buy 15 of the 20 clips on a cheap 720p model ($11.25) and 5 on Seedance 2.5 ($17.33), for $28.58 total.

Remember that the cost of stitching is small by comparison. A two-minute Timeline render is $0.20.

What Sume does and does not do

Sume routes the request to the model you name or picks one with sume/auto, reserves the billable amount on submit and returns hosted clips. It does not decide which shots are good. Use the limits in the catalog, not the planning numbers here, when you script a call.

• Set resolution explicitly; it drives price on most models.
• Keep duration inside the model's range.
• Price retakes, not just the first take.

Rerun the arithmetic

The same line items as a short script, so you can change a quantity and see the total. It is plain arithmetic on the rates in this post; it does not call the Sume API.

# Sume billable USD rates, read 2026-10-03. Edit the quantities.
lines = [
    ('Wan 3.0 720p clip seconds, 15 clips', 90, 0.125),
    ('Seedance 2.5 720p clip seconds, 5 clips', 30, 0.5778),
]
total = 0.0
for label, qty, unit in lines:
    cost = qty * unit
    total += cost
    print(f"{label:<46}{qty:>8g} x {unit:<10g} = ${cost:,.2f}")
print(f"{'Total':<46}{'':>8}   {'':<10}   ${total:,.2f}")