H3 Max Recast vs Genjutsu: which person swap to call on Sume
Sume lists two person-swap video rows. Recast: 1-4 people in a 5-30 s clip at 768p or 1080p. Genjutsu: 1-8 images at 480p or 720p. How to choose.

Use H3 Max Recast when you want the people in an existing clip replaced by people from photos while the footage stays as it is, at 768p or 1080p. Use Higgsfield Genjutsu Motion Transfer when you want 1 to 8 reference images driven by the motion of a source video at 480p or 720p. Both take one source video and both are video-to-video rows in Sume's catalog.
The differences that decide the call are the clip length, the resolution, how many photos you can send, whether a prompt is needed, and availability: Genjutsu is listed only when its provider is configured, while Recast is listed in the standard video catalogs. Everything about Sume's rows below comes from the Video Router and Video generation docs; the Recast model description is from fal's H3 Max Recast page, read 2026-10-02.
How do the two rows differ on paper?
Read the table as Sume exposes them, not as each vendor markets them. Fields not in the Sume docs are left out rather than guessed. Price per second for Recast is fal's list, $0.30 at 768p and $0.45 at 1080p, with Sume billing list times 1.25 on every model. Genjutsu's rate in Sume's pricing fixtures is $0.318 per second at 480p and $0.681 per second at 720p before the same margin.
| h3-max-recast | higgsfield-genjutsu | |
|---|---|---|
| Source video | One, 5 to 30 s, no shot over 15 s | One, 4 to 30 s |
| Photos | 1 to 4, one per person | 1 to 8 reference images |
| Resolution | 768p (default) or 1080p | 480p (default) or 720p |
| Prompt | Optional | Required (non-empty) |
| Audio references | Not accepted | Not accepted |
| Listed when | Standard video catalogs | Only when its provider is configured |
| Price per second (before list × 1.25) | $0.30 (768p), $0.45 (1080p) | $0.318 (480p), $0.681 (720p) |
Which one keeps the original scene?
Recast is described by fal as recasting the people while preserving the source motion, camera, cuts and audio. That makes it the right pick for an ad or a short that is already cut: you keep the edit and change who is in it.
Genjutsu is called Motion Transfer in Sume's docs: the source video supplies motion and you supply reference images. The docs I read do not say it preserves the source's cuts or audio, so I would not rely on that. If audio matters, check the result with a video inspect and probe.has_audio before you promise it to anyone.
What about length and resolution?
Recast takes up to 30 seconds and, in return, gives you 1080p. Genjutsu starts shorter, at 4 seconds, and tops out at 720p. For social formats at 1080p, Recast is the only one of the two that gets there. Genjutsu's default is 480p, which suits a quick draft at a lower per-second rate than 720p.
Both need the source duration up front. Genjutsu requires duration to be the source length (4 to 30); Recast needs the inspected length rounded up, and refuses anything outside 5 to 30. In both cases, probe first with video inspect.
How do you decide in practice?
Ask three questions. Is the clip already edited and should stay that way? Pick Recast. Do you have more than four people's worth of references, or want a draft at 480p? Pick Genjutsu. Is Genjutsu missing from your catalog? It is only listed when its provider is configured, so check GET /v1/video-router/models and fall back to Recast.
Neither row accepts audio references, and Recast refuses first and last frames outright. If you need a brand new clip with a character you describe, that is a different job: see which Sume video models accept a reference video for the generation route.
Sources
Related posts
More in Comparisons
- Hedra's developer platform: API, SDK, CLI and MCP vs Sume
Hedra opened its models through an API, SDKs, a CLI and MCP on August 4, 2026. A map of what each surface covers and what Sume offers for avatar work.
- Hedra pricing plans in credits vs Sume per-second avatar video
Hedra sells credits: $20, $50 or $100 a month plus a $75 Teams plan, with no per-minute avatar cost on the page. Sume publishes a rate per second instead.
- HeyGen break tag: 5-second pause max, and a longer pause on Sume
HeyGen's professional voice clones accept a break tag with a 5-second cap per pause. On Sume an avatar video pause is a silence scene, up to 60 seconds.
- HeyGen free plan limits vs paying per second on Sume
HeyGen's free plan lists 3 videos a month of up to 1 minute. Here is what that means for a test project, and how Sume's per-second billing compares.
Written by Sume