Change a person in a video with AI: Sume Recast price, not free
Changing a person in a video with Sume's H3 Max Recast is paid per source second: $0.375 at 768p, $0.5625 at 1080p. Costs for 5 to 30 second clips.
People searching for a free way to change a person in a video will not find one in Sume's Recast route: h3-max-recast is paid per second of source video. fal lists $0.30 a second at 768p and $0.45 at 1080p; Sume bills that list price times 1.25, which is $0.375 and $0.5625 a second. A 10 second clip at 768p is $3.75.
Sume's docs do not describe a free tier for this model. Workspaces draw from their plan's included Gen$ and balance, and a request the balance cannot cover is refused with 402 insufficient_credits before work starts. The numbers below are the whole bill.
What does a Recast clip cost on Sume?
The rate card is per output second, and the output keeps the source length. Sume rounds the source length up to whole seconds, so a 12.2 second clip bills as 13. Reference photos add nothing. Sume's fixtures round the billed amount to cents, and the whole-cent figures below follow that.
- Resolution is the only price lever besides length. Sume defaults to 768p; fal's own default is 1080p, so a fal playground number will not match an unlabelled Sume call.
- Five seconds is the minimum and 30 the maximum; anything outside is rejected before it can bill.
- Number of people (1 to 4) and number of photos do not change the price.
| Source length | fal 768p | Sume 768p | fal 1080p | Sume 1080p |
|---|---|---|---|---|
| 5 s | $1.50 | $1.88 | $2.25 | $2.81 |
| 10 s | $3.00 | $3.75 | $4.50 | $5.63 |
| 15 s | $4.50 | $5.63 | $6.75 | $8.44 |
| 20 s | $6.00 | $7.50 | $9.00 | $11.25 |
| 30 s | $9.00 | $11.25 | $13.50 | $16.88 |
What do real jobs cost?
A 14.2 second ad bills as 15 seconds: $5.63 at 768p and $8.44 at 1080p. Four creator variants of that ad at 768p are four separate jobs, 4 x $5.625, or $22.50 in exact dollars before cent rounding. Twenty 10 second clips at 768p come to $75.00.
Against fal's own list the Sume premium is the 25 percent house margin: $0.075 a second at 768p, or 75 cents on a 10 second clip. What you get for it is the Sume job record: one API key, an Idempotency-Key on every submit, status and result endpoints, webhooks and the hold-and-refund behaviour described above. Whether that is worth it depends on whether you already run a fal account.
What happens to the money if the job fails?
At submit, Sume reserves the estimated amount. Per the admission docs, completion captures the reserved usage, and failed jobs and failed queue admission release or refund the reservation where applicable. Cancelling is possible only before generation starts; once it starts, cancel returns 409 job_generation_already_started and the job finishes or fails normally.
That is why the duration field matters on Recast. It is reserve-only: the output keeps the source length, but Sume needs your inspected source seconds to quote the hold. Send the wrong number and you either get a rejection (outside 5 to 30) or a hold that does not match the clip.
How do I keep the bill down?
Trim before you submit. Recast changes the people for the whole source, so every second you do not need is money. If only a 6 second beat has the person to change, cut that beat and recast it; a trimmed clip must still be at least 5 seconds. For footage past 30 seconds, trim, recast each part and rejoin.
Test the mapping on the cheapest valid clip: 5 seconds at 768p is $1.88 (exactly $1.875). Only the final pass is worth 1080p. And make sure you have permission for every face you submit; Sume's Terms of Service put that responsibility on the person submitting the likeness.
Is there a cheaper way to swap a person on Sume?
Sume lists a second person-swap row, higgsfield-genjutsu, which takes one source video and 1 to 8 reference images at 480p or 720p for 4 to 30 seconds, and is listed only when its provider is configured. It is a Motion Transfer model, not the same job. Sume's cost comparison of the two puts Recast at the low end of the video-to-video range, and its post on choosing between them covers when each is the right tool. If you need the swap at 1080p, Recast is the row that offers it.
There is also a cost that never appears on the invoice: re-doing a clip that should have been checked first. Confirm that the source is 5 to 30 seconds, that no single shot runs past 15, and that you have the rights to every face, before you press submit.
Where do I read the live price?
Sume's catalog is the source of truth, not a blog table. GET /v1/video-router/models/h3-max-recast returns the per-second list price by resolution and the billable formula. fal can change its list at any time, so compare against its model page before you budget a large batch.
curl https://api.sume.com/v1/video-router/models/h3-max-recast \
-H "Authorization: Bearer $SUME_API_KEY"Sources
Related posts
More in Pricing
- Cheap AI Video Drafts Before the Final Render: 360p, 480p, Grok Rates
Draft a shot for pennies: Sume rates for 360p Omni, 480p Wan and H3, and Grok image-to-video, with the draft-to-final cost of a 5-second shot.
- Cheapest 10-second 1080p AI video on Sume: five models ranked
A 10-second 1080p clip costs $1.40 on Kling 3 with audio off and $2.50 on Wan 3.0. The ranking by Sume price, with what each model does at that size.
- Cheapest AI image edit per reference slot on the Sume catalog
Reference slots per dollar for every edit-capable Sume image model: Qwen Image and Grok Imagine lead at 400, ChatGPT Image 2.5 offers the most slots at 16.
- Cheapest AI person swap on Sume: the 5-second Recast floor
The lowest price for an H3 Max Recast job on Sume is $1.875 (5 s at 768p). Shorter clips are rejected. Floor table and a cheap way to test a face first.
Written by Sume