Most writing about video models is about the output. This site is about the other end — the gate the request has to pass before a GPU is ever involved.
That gate is where the majority of failed jobs actually die, and it is the cheapest place for them to die, because nothing has been generated and nothing should be billed. It is also where the error messages are least useful, because a validator that rejects a request rarely explains which of four different budgets you exceeded.
The worked example throughout is Alibaba's Wan 3.0, read against Model Studio's own API reference. The shapes generalise: two mutually exclusive input families, limits that are totals rather than per-item, a page count that is not a file size, and a handful of format rules that produce rejections which read as mysterious until you have seen each one once.
The most common rejection in Wan 3.0 arrives as this, and it arrives before any compute has run:
Here is the accepted image list for Wan 3.0, as published: JPEG, JPG, PNG, BMP, WEBP. Sides between 240 and 8000 pixels, aspect ratio within 8:1, up to 20 MB.
This is the bug that ships. It ships because it passes every test you would naturally write.
The document route on Wan 3.0 accepts one file, up to 50 pages, up to 100 MB, in docx, doc, xlsx, xls, pptx, ppt, pdf, txt, key, pages, numbers and md.
Every image going into Wan 3.0 has to sit inside three constraints at once:
An ingest path for a video model has four categories of refusal, and they need four different sentences. Most implementations write one.
Everything else on this site is about files. This page is about the text box, because it behaves like an upload path and almost nobody validates it like one.