H3 Max Recast 768p vs 1080p: price per second and per clip

H3 Max Recast's 1080p tier costs 50 percent more per second than 768p: fal's model page bills output video at $0.30 per second for 768p and $0.45 per second for 1080p, a 50 percent step up. On Sume the same tiers bill at the vendor list times 1.25, which is $0.375 per second at 768p and $0.5625 per second at 1080p, so a 10-second clip is $3.75 at 768p and $5.625 at 1080p.

Sume's default is 768p, so a request that omits resolution is priced at the lower rate. This post gives the per-clip numbers for the whole 5 to 30 second window, shows what the extra 1080p dollars buy you in cost terms, and explains how the amount is held and settled. Vendor numbers come from fal's model page, read 2026-10-03; Sume numbers come from the Video Router docs and the pricing constants in Sume's repository.

What does fal list for H3 Max Recast?

fal's page for minimax/h3-max/recast says the model is billed by output video duration: $0.30 per second at 768p and $0.45 per second at 1080p, with reference images included at no extra cost. It recasts people in a video using reference photos while keeping the source motion, camera work, cuts and audio. The page lists mp4, mov, webm, m4v and gif video inputs.

A news digest on October 1 reported a 5 to 30 second clip window for the model. That window is the one Sume enforces, but treat it as reported: only the price lines above come from the vendor page itself.

How does Sume turn that into a per-second price?

The Video Router docs state that Sume bills list times 1.25 on every catalog model, and that the amount is reserved when you submit. For h3-max-recast, the output keeps the source clip's length, so the length you send as duration is the number of billed seconds, rounded up to a whole second. The accepted range is 5 to 30 seconds, and the resolution must be 768p or 1080p.

Multiplying out: $0.30 x 1.25 = $0.375 and $0.45 x 1.25 = $0.5625. The gap between the tiers on Sume is $0.1875 per second, which is $1.875 on a 10-second clip and $5.625 on a 30-second clip.

What does one clip cost at each length?

Every figure below is the clip length in whole seconds times the Sume per-second rate. They are list-price arithmetic, not a measurement of any run.

Is 1080p worth the extra $0.1875 a second?

That depends on where the clip ends up, and nothing in Sume's docs says 1080p changes the recast itself. The resolution is an output size, so the cost question is simple: if the clip is going into a vertical social post, the 768p tier is the cheaper default and is the one Sume sends unless you override it. If the clip will be cut into a 1080p timeline, you pay the premium once on each clip you recast.

A practical way to keep the bill down is to test the swap at 768p on the shortest representative clip, check the faces, and only re-run at 1080p for the approved takes. Because a resubmission is a new paid job, test at the lower tier first.

• 768p is the Sume default, so omit resolution for the cheaper tier.
• 1080p adds $0.1875 per output second over 768p on Sume.
• The source duration is rounded up, so a 10.2 second clip bills as 11 seconds.
• Reference photos add no extra line; fal and Sume both price on output seconds.

How is the amount held, and what if the job fails?

At submit time Sume reserves the estimated amount from your USD balance. If the workspace cannot cover it, the call is refused with 402 insufficient_credits before any provider work starts. On success the reservation is captured; on a failed job, or a cancel before generation starts, the reservation is released or refunded. The Usage docs list the three ledger statuses as reserved, captured and refunded, and GET /v1/usage?job_id=... sums one job's money.

A 30 second 1080p clip holds $16.875 at once, so check GET /v1/balance before a batch. Here is a submit that sets every cost-relevant field explicitly:

A sensible workflow follows from the numbers. Run the 768p default while you test the prompt and the reference, since each second costs $0.375 instead of $0.5625, and move to 1080p only for the take you plan to publish. Remember that the source length is rounded up to a whole second before either rate applies, so trim the source first.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: recast-768-demo-001" \
  -d '{
    "model": "h3-max-recast",
    "video_url": "https://example.com/source.mp4",
    "reference_image_urls": ["https://example.com/person-a.jpg"],
    "resolution": "768p",
    "duration": 10,
    "mode": "async"
  }'