Swap the presenter per market: one ad, two Recast jobs on Sume
Recast keeps the motion, camera, cuts and sound of one source ad and replaces the people. How to make one version per market, the limits and the price.

To show a different presenter in each market, shoot the ad once and run one H3 Max Recast job per market, each with a different set of reference photos. Recast replaces the primary people in the source video with the people in your photos and keeps the motion, the camera, the cuts and the sound of the source. So the ad is the same in every version and only the person changes.
The sound stays too, which has a consequence that is easy to miss: if the voice speaks English in the source, every version speaks English. Recast changes who is on screen and not what is said.
What one job takes
Recast is the h3-max-recast id on the Video Router. The inputs and limits are fixed by the catalog. Because the route reads the length of your source, the cheapest version of a test is a short cut of the master, five to eight seconds long, with the presenter in frame. Check that the face swap holds up on that cut before you pay for the whole ad in every market.
| Item | Value |
|---|---|
| Source | video_url, 5 to 30 seconds |
| Shots | No single shot longer than 15 seconds |
| People | 1 to 4 reference_image_urls, one photo per new person |
| Prompt | Optional, up to 2000 characters |
| Resolution | 768p default, or 1080p |
| Output | Keeps source motion, camera, cuts and sound |
| Length | The inspected length of the source, rounded up |
A per-market workflow
Cut the master ad so that no shot passes 15 seconds, and if the ad is longer than 30 seconds, split it into sections that are each within the window. Collect one clear photo of the presenter for each market, and submit one job per market with the same source. Keep the prompt short and use it only to say what to preserve, such as the wardrobe or the setting, because the source already carries the action.
Think about what is truly local in each market and what is not. The presenter, and with a prompt the wardrobe, can change by market. The set, the product, the camera work and the voice cannot, because they come from the source. If the product packaging or the on-screen text must also differ by market, that belongs in the master shoot or in a later edit, and not in the Recast job. Deciding this before the shoot saves a re-shoot, because the master has to be right for every market it will serve.
Run the jobs under separate idempotency keys that name the market, so that a retry for one market never collides with another. When the results come back, review each at full size before the version goes into a campaign, since each is a separate render.
What it costs for a set of markets
The rate is $0.375 per second at 768p and $0.5625 per second at 1080p, which is the list $0.30 and $0.45 times 1.25, applied to the source length rounded up. The price scales with the number of markets and nothing is shared between jobs.
| Markets | 768p | 1080p |
|---|---|---|
| 1 | $5.625 | $8.4375 |
| 2 | $11.25 | $16.875 |
| 4 | $22.50 | $33.75 |
What stays the same, and what does not
Rights matter here too. Use photos of people who have agreed to appear in the ad, and keep the written agreement with the asset. The tool changes pixels, but the permission has to come from the person.
Name the photos and the jobs by market in your own records, and keep the reference photos next to the source file. When a campaign is renewed a year later, the same inputs reproduce the same version, and a new market is one more job against the same master.
Because the camera and the cuts are kept, the framing in each version matches the master, so the presenter's photo should have a similar framing and a similar angle to the person in the source. The sound is the master's sound. If a market needs another language, record that source version first, and then recast that one. See also whether Recast keeps the original audio and swapping two people in one clip.
Sources
Related posts
More in Use cases
- SynthID on every Gemini TTS clip: keep the original file for ad review
Google marks every Gemini 3.8 TTS clip with SynthID and C2PA credentials. What that means for ad voiceovers, and what Sume docs do and do not say.
- 10 static ad variants in one Sume image request: n and its cap
The Sume Image API takes n from 1 to 10, but each model has its own cap. Read it from the catalog; use 4:5 on Nano Banana or 1088 x 1360 on GPT for feed ads.
- Test the hardest 5 seconds first: trim, then Recast before the full 30
Before paying $11.25 to recast 30 seconds at 768p, trim the worst 5 seconds and recast just that for $1.895. What the test can and cannot tell you.
- Threads AI Info label: 150 million views in October 2024, how visible?
Meta's snapshot shows Threads AI Info labels with 150M+ views and 480k+ user actions in one 2024 month. What it says about visibility for AI video on Threads.
Written by Sume