Which model does this shot belong on?
Tick what the shot actually does. The answer is the routing rule we run on every channel, including the two exceptions that cost us the most to find.
The rule this encodes
| Motion | Model | What belongs here |
|---|---|---|
| Tier 1 — low or no motion | Kling 3.0 | Calm, observational, aftermath, a still subject, a soft drift. |
| Tier 2 — some motion | Seedance 2.0 fast | Walking, tracking, a zoom, a pan. |
| Tier 3 — hard motion | Seedance 2.0 std | Fighting, sprinting, a chase, an explosion, a collapse. |
It is a cost rule, not a taste rule
Running a static shot on the expensive tier is pure waste — the output is not better, it is only dearer. Running a hard-motion shot on the cheap tier is worse than waste, because the re-order costs more than the correct order would have. Most episodes should come out majority Tier 1. If yours does not, the shot list is the thing to look at.
The two exceptions worth knowing
Lip-sync overrides everything. A singing or speaking face is nearly still, so every motion heuristic files it as Tier 1 — and then the mouth does not move. This is not a tuning problem: the cheap tier has no audio input at all. Declare lip-sync in the script so the router cannot mis-file it.
Resolution is set by format, not by model. Shorts are 720p, mid and long-form are 1080p. Fast mode cannot render 1080p, so a Tier 2 shot inside a long-form episode is promoted to the standard model to hold the resolution rule. In shorts there is no promotion.