Why a batch beats one expensive first attempt
Video generation is stochastic. The same source and character can produce different hands, edges, object contact, and first frames. One output tells you whether the route can work; a controlled batch tells you how often it works for this shot.
Start with four outputs when you are qualifying a new source. Use eight when the motion has already passed a basic test and you need enough candidates to review. Twelve or twenty outputs make more sense only after the source, character pose, and route are stable.
Keep the experiment constant
Use the same source file, character image, duration, resolution, consent record, and model route across the batch. The provider may vary its internal seed, but the commercial question stays clear: which output best preserves this exact shot?
If you change character, wardrobe, language, and camera treatment at the same time, you no longer have a batch test. You have several experiments mixed into one bill.
Quote the whole batch before reserving credits
The console should show output count, seconds per output, total generated seconds, per-output credits, and total credits before the job is submitted. A 20-output button without a total is a spending trap, not a production tool.
ClipRecast reserves the quoted batch amount before provider work begins. That keeps two tabs from spending the same balance and prevents a batch from stopping halfway because only some outputs were funded.
Reconcile each output independently
A batch can contain successful, failed, and still-processing outputs at the same time. The result view needs a separate status, preview, and authenticated download for every item instead of collapsing the entire batch into one vague job state.
Eligible credits should return for confirmed technical failures at the output level. Valid outputs that simply rank lower in your creative review are still completed generations, not provider failures.
Review with a repeatable scorecard
Check identity consistency, hands, feet, object contact, garment edges, background preservation, camera motion, and the opening and closing frames. Mark the reason an output failed instead of relying on an unrecorded impression.
The scorecard is what turns a larger purchase into learning. It shows whether the next spend should use more outputs, a better character reference, another source crop, or a different provider route.
Scale only the winning setup
When one source-and-character pair is stable, repeat the same setup for a larger batch or move to the next character. Preserve the original quote and results so each campaign can be compared by cost per usable output, not just total files generated.
Batch size is a production control. The right size is the smallest group that gives the team enough usable candidates to publish and learn from the test.
This guide describes a product workflow and general information; it is not legal advice.