H3 Max Recast 768p or 1080p: draft cheap, publish in HD
Sume bills H3 Max Recast at $0.375 a second at 768p and $0.5625 at 1080p. Budget a 768p test pass and one 1080p final, and what each plan costs.

For H3 Max Recast, 1080p costs 50 percent more per second than 768p: fal lists $0.45 against $0.30, and Sume bills those times 1.25, so $0.5625 against $0.375. Use 768p for tests and 1080p for the take you publish. Sume defaults to 768p when you leave resolution out, and fal's own default is the opposite, 1080P, so an unlabelled call will not match a fal playground price.
What is the real price gap?
The gap grows linearly with source length, so it is small on a test slice and meaningful on a 30 second clip:
| Source length | 768p | 1080p | Extra for 1080p |
|---|---|---|---|
| 5 s | $1.88 | $2.81 | $0.94 |
| 10 s | $3.75 | $5.63 | $1.88 |
| 20 s | $7.50 | $11.25 | $3.75 |
| 30 s | $11.25 | $16.88 | $5.63 |
Can I trust a 768p draft to predict the 1080p result?
Not exactly, and this is the catch in the draft-then-final workflow. Sume has no seed on this model, so a 1080p run is a new take rather than a higher-resolution copy of your 768p draft. What a draft does tell you is whether the mapping works: that the right photo landed on the right person, in every shot, and that the clip is acceptable to the model at all.
Sume does not publish how the two resolutions are produced for Recast, so do not assume the 768p result is a downscale of the 1080p one. For minimax-h3-max text and image video Sume states that its 1080p is a latent refinement from native 768p; that statement is about that model id and is not claimed for Recast.
What should a 768p test tell me?
A test pass answers specific questions, so write them down before you spend:
- Did each photo land on the intended person, in every shot where that person appears?
- Do two people who cross or swap sides keep their identities?
- Is the source audio present and in sync, as fal says it is retained?
- Do the cuts and camera moves match the source, which fal says are preserved?
- Is the clip accepted at all, with no failure from a long shot or an unusual source?
How should I budget a project?
Three plans for a 10 second clip, each with the arithmetic:
- One shot at 1080p: $5.63. Cheapest if the first take is fine, which you cannot count on.
- Two 768p tests and one 1080p final: 2 x $3.75 + $5.625 = $13.125, shown as $13.13. A sensible default when mapping is uncertain.
- Four 768p tests and no 1080p: $15.00. Right when the deliverable is a social post that will be compressed anyway.
When does 1080p earn its price?
When the finished clip will be shown larger than a phone feed, when the source itself is 1080p and softness in the person would stand out against sharp surroundings, or when it is going to be cropped in editing. When the destination is a 9:16 story or a short cut-down, 768p is often enough, but that is a judgement about your channel, not a Sume claim.
Note also that fal's price is per second of output, not per pixel. A 5 second clip at 1080p ($2.81) costs less than a 10 second clip at 768p ($3.75), so cutting length is the stronger lever than dropping resolution.
What about the cost of an upscale afterwards?
A common thought is to recast at 768p and enlarge the result later. Sume's pricing for Recast covers only the two resolutions, and its docs describe no upscale step for this model, so any enlargement would be a separate tool and a separate cost you would have to price yourself. For a clip that must ship in HD, it is simpler to pay for the 1080p take.
Whatever you choose, quote it to a client from the catalog, not from memory: GET /v1/video-router/models/h3-max-recast returns the per-second list price for each resolution and the billable formula, so a budget sheet can be rebuilt from the live numbers whenever fal changes its list.
How do I pin the resolution in a request?
Send it explicitly every time. Sume accepts only 768p and 1080p on this model and rejects other values such as 480p before any paid work starts. The field is optional, but pinning it removes the surprise when a defaults page changes.
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-final-1080-001" \
-d '{"model":"h3-max-recast","video_url":"https://example.com/original.mp4",
"reference_image_urls":["https://example.com/person.png"],
"resolution":"1080p","duration":10,"mode":"async"}'Sources
Related posts
More in Pricing
- H3 Max Recast 768p vs 1080p: price per second and per clip
fal lists H3 Max Recast at $0.30 a second for 768p and $0.45 for 1080p. On Sume the same two tiers cost $0.375 and $0.5625 a second; per-clip math for 5-30 s.
- Ideogram 4 self-hosting: $300 a month licence vs per-image API
Ideogram's self-serve licence is $300 a month for up to 10,000 images, before GPU time. Where it beats a per-image API such as Sume's, and where it does not.
- Kling 3.0 Motion Control price per second and a 30-second clip
fal lists Kling 3.0 Standard Motion Control at $0.126 a second. On Sume it is $0.1575 a second, rounded up per second: $4.725 for the 30-second maximum.
- MAI-Transcribe-2-Streaming at $0.54 an hour vs Sume STT per hour
Microsoft lists $0.54 per audio hour through year end. Sume STT is $0.01 a minute ($0.60 an hour) as 10-minute async jobs. The cost at 1 to 1,000 hours.
Written by Sume