What this site is
BestVideoGenerators is a head-to-head comparison reference for AI video generation models. Every comparison page is built from a shared data sheet in which each table cell carries a value, a source URL and a checked date — or the words “not published”. We would rather show a gap than invent a number.
Sourcing rules
- Official sources only. Every value in a comparison or spec table comes from the vendor's own pages (pricing pages, API docs, launch posts, help-center articles) or, for rankings, from the leaderboard or benchmark publisher itself. Each value is stored as value + source URL + checked date and shown with a numbered [n] citation that links to the Evidence list at the bottom of the page.
- No public data means no number. If we could not find a value on an official page, the cell says “No public data”. Third-party trackers, resellers, press articles and Wikipedia are used only as leads for research, never as the published value.
- Licensed data stays out. We do not republish rankings whose terms require permission or a data licence.
- Nothing fabricated. No invented benchmarks, scores, votes, “we tested” claims or star ratings.
How winners are decided
Each comparison has category tables (pricing, output specs, features, access and public leaderboards). A row is won by the lower price, the longer clip, the higher resolution or frame rate, more reference inputs, or a supported feature. Prices of a discontinued product are shown as historical and cannot win. LMArena scores whose ± confidence intervals overlap count as a tie. A category goes to the model that wins more rows; equal wins is a tie, and a category with no comparable rows is marked “Depends”. The quick verdict adds one editorial sentence on top of that tally.
Platform rankings and the “Model variety” category
The platform hub carries an overall ranking. It is editorial: it orders the platforms by model breadth, pricing transparency and credit flexibility, and it is labelled as editorial order on the page. The “Model variety” category inside each platform comparison is computed from what each platform itself publishes: its own model count where it states one, otherwise the distinct AI video and image models it names on its own pages (a floor, not a ceiling). Counts that bundle audio, music, stock or LLM models are not comparable and are not used.
Hands-on tests and placeholders
We have not yet run same-prompt generation tests ourselves, and no page claims otherwise. There are no example, placeholder or “pending” ratings anywhere on this site: a cell either carries a sourced value with its [n] citation and checked date, or it says “No public data”. Open verification items are tracked in an internal ledger that is not published.
How pairs are chosen
We build a comparison page only when the pair has genuine, distinct analysis (a spec overlap and a real user decision between them). We do not generate hundreds of programmatic “X vs Y” permutations; four matchups cover the models indexed here. Canonical pair order is fixed and reverse slugs 301 to the canonical page.
Product links
Where a page recommends AIContentDrop, the recommendation is an editorial one and its links are plain citations like every other source on the page.
Update cadence
Pricing and limits: re-checked monthly and on major vendor announcements. Model status (releases, discontinuations): on release. Every page shows a visible “Last verified” date; anything you spot that is out of date belongs on the corrections page.