Nothing stored between calls, against a mechanism with a name

One page documents one to seven reference images per call and no object surviving the request. The other names Hyperframe as what locks a face, wardrobe and voice across every episode. These are the two ends of this column, and the fork is who keeps the artefact. As of 2026-09-22.

Fig. 1 The rows are the three questions a production asks on the morning a lead arrives looking like a relative of herself.

Vidu and Pexo on the cross-episode cast, point by point. Read 2026-09-12 and 2026-09-22.
Point at issue	Vidu	Pexo
Where the character lives	In the caller's own files, resent on every request	Inside the product, under a name
What survives a request	Nothing documented	The lock, stated for every episode
Who repairs a drift	The caller, by fixing what they send	The vendor, by whatever Hyperframe does

Inclusion rule. Both cells are graded documented. The contrast is not about strength of evidence but about where the artefact is held, which the two pages answer incompatibly. Order. The page that stores nothing first, then the page that names a store.

1The two arrangements fail in different weeks

Resending the same images works perfectly on day one and degrades slowly. Somebody renames a file, a second operator joins and uses a slightly different crop, a shot is regenerated in a hurry without the references attached. None of that is a product defect and all of it shows up on screen.

A named lock inside the product cannot drift that way, and it can fail in a way the customer cannot see or fix. The support conversation is about somebody else's machinery, and the production waits.

2Published input limits point at who the page was written for

A minimum pixel size and an allowed aspect-ratio range belong to a developer integrating against an interface. A sentence about a face, a wardrobe and a voice across every episode belongs to a producer deciding what to buy.

Neither audience is better served in the abstract. A studio with engineers gains from the first; a studio with a schedule and no engineers gains from the second, and buying the wrong one of the two is a staffing mistake rather than a technical one.

3What neither page settles

No sentence on either page says how near the ninth episode's face lands to the first. That needs a production run, a shot list nobody has published and a ruling on how near counts as near enough, and it is the question both readers actually have.

So the contrast is about the arrangement, not the result. It tells a buyer which afternoon goes on repairing a drift and whose afternoon it is.

4The wording both sides published

Cross-episode cast
One to seven reference images per call, used so the model generates a video with consistent subjectsper call rather than per characterVidu, reference-to-video API reference / recorded 2026-09-12
Cross-episode cast
Hyperframe is named as what locks a character's face, wardrobe and voice across every episode, and the identity of a product or brand in every shota named mechanismPexo, micro drama page / recorded 2026-09-22

5The same column, read for a third tool

On the cross-episode cast, SceneMixer's published page puts it in one line: A full-body sheet per character, made while the script is analysed. Its reading here, A reference sheet made once and cropped into every segment, sets out what that line settles for a series and what it leaves to a production run.

6Sources

Read from the reference-to-video API reference at platform.vidu.com and the micro drama page at pexo.ai, checked 2026-09-12 and 2026-09-22 respectively. The full column is on cross-episode cast; the single readings behind this contrast are Vidu and Pexo. Other contrasts filed here: LTX Studio and Novi AI, Google Flow and PopShort AI, Medeo and SceneMixer.