Replace the presenter in a 3-minute training video with H3 Max Recast
Recast takes 5 to 30 seconds per job, so a 3-minute video is six chunks. Steps, the 768p cost, and why the original voice track becomes the timeline spine.
To replace the presenter in a 3-minute training video with H3 Max Recast, cut the source into six 30-second chunks, recast each with the new presenter's photo, and join them on a timeline using the original voice track as the audio spine. At 768p the six Recast jobs cost $67.50 and the trims and join add about $0.43.
Why six chunks
Recast accepts 5 to 30 seconds. The duration equals the source length, and the prompt is optional. A 180-second video therefore needs six jobs. Video trim cuts each piece for $0.02 per job, with start plus duration and a precision of exact (frame-accurate re-encode) or keyframe (stream copy). Use exact, so the chunk edges land where you asked.
Cost
Recast list is $0.30 per second at 768p and $0.45 at 1080p. With the 1.25 multiplier, Sume bills $0.375 and $0.5625.
| Step | Count | Rate | Total |
|---|---|---|---|
| Video trim | 6 | $0.02 per job | $0.12 |
| Recast 768p, 30 s each | 6 | $11.25 per job | $67.50 |
| Audio detach | 1 | $0.01 per job | $0.01 |
| Timeline render, 3 minutes | 1 | $0.10 per minute | $0.30 |
| Total at 768p | $67.93 |
Why the voice is the spine
Recast replaces the people on screen. This plan does not rely on the source audio surviving the job. Detach the original track once with audio detach and use it as audio.url with duration_seconds: 180 in the Timeline request. The result then keeps the original narration across all six chunk boundaries. If the new presenter is meant to speak in a different voice, that is a separate lip-sync problem; see the talking-photo comparison.
The join
In video[], list the six recast clips with start at 0, 30, 60, 90, 120 and 150, each with duration: 30. Every URL must be a media.sume.com artifact. Keep transitions out of it, or use a short fade of 0.25 seconds on slots after the first. Run POST /v1/timeline-1.0/plan first; it is unbilled and returns the billable minutes.
Checking the result
Extract a still from the start of each recast chunk and one from the end of the one before it. The two should show the new presenter in a matching position and outfit. If the face changes between chunks, use the same photo for all six and keep the prompt empty or identical across them. After the join, run video inspect with frames: false and check the probe's duration_seconds against 180.
Budget for one redo. A single bad chunk costs $11.25 to run again at 768p, and a full redo of the video costs $67.50.
Chunk edges and the last piece
A source that is not a multiple of 30 seconds ends with a short chunk, and Recast's floor is 5 seconds, so plan the cuts so that the last piece is at least 5 seconds. Cut at sentence boundaries in the voice track, not at a fixed 30, when you can. Test the hardest 5 seconds before the full run: recast video longer than 30 seconds covers the same split in more detail.
Sources
Related posts
More in Use cases
- Replace the view outside a window in an interior photo
Swap the scene seen through a window and keep the curtains, frame and room, then relight in a second pass. Ideogram 4.5 on Sume, with prompts and costs.
- How do I make a Thanksgiving takeout menu video for my restaurant?
Animate two of your own food photos with Gemini Omni Flash 1.1, add a 380-character voiceover and burn the order-by date in: $1.945 on Sume, in 12 seconds.
- Restore a scratched family photo in three staged edits
Fix creases, then scratches, then fading in separate Ideogram 4.5 passes. Three low-quality passes cost $0.1125 on Sume; a high final pass brings it to $0.35.
- Re-upload an old vertical video: is it a Short or long-form?
YouTube's page says square or vertical uploads up to three minutes after October 15, 2024 are Shorts. What it means for a re-upload, and how to check the file.
Written by Sume