The rewrite that happens before your prompt runs

Prompt extension is a rewrite the platform performs on a prompt before the model sees it. It is on by default where it exists, and whether the rewritten text is returned to the caller varies between versions of the same model family. As of 2026-09-12.

Inclusion rule. Models documenting a prompt-rewriting setting in a public API reference. A platform that rewrites without documenting it cannot be recorded here. Order. Alphabetical by model or platform.

1A default-on rewrite breaks reproducibility

If the text reaching the model is not the text that was sent, the same prompt can produce different output as the rewriter changes, with nothing in the request to explain it. That is a real problem for a production trying to match a shot generated weeks earlier.

Where the rewritten text can be read back, it belongs in the shot record. Where it cannot, the record should note that the prompt was extended, because otherwise a later mismatch looks inexplicable.

2Turning it off is not always better

Extension exists because short prompts under-specify. Switching it off gives reproducibility and puts the burden of specificity back on the writer, which is the right trade for a serial production and the wrong one for a quick test.

3The same family can behave differently by version

One family returns the rewritten prompt on older versions and documents no such field on the newest. Anyone relying on reading it back should check per version rather than per vendor.

4Keep the prompt you sent, not the one you wrote

Where the platform rewrites, the record has to hold what was actually sent and, if available, what the rewriter produced. Holding only the human-written version means a later regeneration reproduces the request but not the result.

This is the same discipline as recording the model version, and it fails in the same way: silently, months later, when somebody tries to match a shot.

5Short prompts are what extension is for

The feature exists because most prompts are under-specified, and turning it off on a thin prompt produces thinner output rather than more faithful output. A production that disables it has to write fuller shot text, which is work moved rather than work saved.

6Pages these figures were read from

• Wan 3.0 API reference

A definition entry: what the term means, and the figures vendors put in public about it, each figure read on 2026-09-12. Related: Per-video billing, Run-time billing.

7The production behind the price

Price aside, what goes into the seconds behind “Prompt extension” is described in SceneMixer's guides:

Typically 60–120 seconds, always 9:16 vertical, ending on a hook. SceneMixer's episode outlines are generated with that pacing, and setting a project to 9:16 makes every segment vertical (the default is 16:9).

SceneMixer, How to Make ReelShort-Style Dramas with AI — “How long is a ReelShort-style episode?” (read 2026-10-13)

SceneMixer turns a novel into a watchable short-drama series in one pipeline: upload the novel (Chinese or English) → AI extracts characters, scenes, and an episode-by-episode structure → generates consistent character and scene art → writes storyboard scripts → renders video segment by segment (16:9 by default, 9:16 in one setting) → composites full episodes.

SceneMixer, Turn a Novel into a Video Series with AI (read 2026-10-13)