H3 Max Recast source rules: 5 to 30 seconds, no shot over 15
Recast swaps the people in a video for 1 to 4 photos. Source: 5 to 30 s, no shot over 15 s, duration set to the inspected length. $0.375 a second at 768p.

The short answer
H3 Max Recast replaces the main people in a source video with the people in 1 to 4 reference photos. The source must run 5 to 30 seconds with no single shot longer than 15 seconds, and the duration you send must be the inspected length in seconds, rounded up. At the default 768p it bills $0.375 a second, so 30 seconds is $11.25.
What Recast does
Recast is video-to-video. The output keeps the source's length, motion, camera, cuts and soundtrack, and only the people change. There is no text-only mode, and aspect_ratio, frame_images, generate_audio, seed and audio references are not accepted. A prompt is optional, up to 2000 characters, and says who becomes whom or what else to keep.
The 15 second shot rule
A source of 5 to 30 seconds must also have no continuous shot longer than 15 seconds, per the catalog. A talking-head clip filmed in one take for 25 seconds is the typical case that breaks it. Cut the clip into pieces under 15 seconds at natural pauses, run each piece, and join the results in a Timeline render.
How to check a source
The route does not detect shots for you. Video inspect returns probe facts and sampled stills but not typed scenes, so look at the stills, or know your own edit points, then cut with video trim. Each trim cut is a range [start, end) from a clip on the media host.
What it costs
Billing is per second, so splitting a 24-second take into two 12-second pieces costs the same as one pass would have.
| Resolution | Per second | 15 s | 30 s |
|---|---|---|---|
| 768p (default) | $0.375 | $5.625 | $11.25 |
| 1080p | $0.5625 | $8.4375 | $16.875 |
Request shape
On /v1/videos, the inputs go in input_references: exactly one video_url and 1 to 4 image_url entries, one photo per new person, left to right by default. The duration is the length you got from inspecting the source, rounded up. Do not guess it.
curl -X POST https://api.sume.com/v1/videos \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: recast-001" \
-d '{
"model": "h3-max-recast",
"prompt": "Replace the presenter with the person in the photo, keep everything else",
"duration": 12,
"resolution": "768p",
"input_references": [
{"type": "video_url", "video_url": {"url": "https://example.com/take.mp4"}},
{"type": "image_url", "image_url": {"url": "https://example.com/new-presenter.png"}}
]
}'Sources
Related posts
More in Models
- H3 or H3 Max? Pick the Sume row by job, with per-clip prices
Both read references and make stereo audio. H3 is $0.075 per second at 768p, H3 Max $0.10 with a 1080p row. A job-by-job guide with 10-second prices.
- higgsfield-genjutsu missing from /v1/videos/models: why
higgsfield-genjutsu is in the Sume video catalog only when its provider is configured. It is Motion Transfer: one video_url plus 1-8 images, 480p or 720p.
- How long can one AI video request be? Sume duration windows
Duration windows differ by model: 2 to 30 seconds on wan-3.0, 4 to 30 on seedance-2.5, 3 to 10 on Gemini Omni Flash 1.1. Read them from the catalog.
- Clips for a 180-second Short: Wan 3.0, Seedance 2.5, Kling 3, Omni
A 3-minute Short needs 6 to 18 clips depending on the model's maximum duration. See the table, then join them with Timeline 1.0; auto chunking does the rest.
Written by Sume