InVideo has moved fast and renamed itself repeatedly while doing it. The core product is now fronted as Agent Two, with Agent Intelligence announced in July 2026, and before that there were earlier generations of the AI video pitch. The help centre carries about a hundred and thirty one articles across eleven collections, the largest being Autopilot with around fifty one and AI twins and avatars with twenty two. That distribution is honest about what people actually use.
The writing is fine and the coverage is broad. The difficulty is that documentation for a product which rebrands its central workflow every eighteen months is always partly archaeological. Work through enough of these articles and you will find guidance describing an interface or a concept that has since been renamed or absorbed into something else, and as a user you cannot easily tell which is which. That is a hard problem and it is one InVideo created for itself by shipping and renaming at this pace.
A separate Studio support site adds another surface to check. The commercial structure is worth flagging because it affects learning directly. The pricing page has no free tier, with plans beginning at Plus and rising through Max, Generative and Elite, all billed annually, and reported figures vary enough between sources that I would not quote them with confidence. What that means practically is you cannot work through the tutorials to find out whether the tool suits you.
You pay, then you learn. For a category where output quality varies enormously by use case, that is a meaningful barrier and it puts more weight on documentation that is not really built to carry it. On substance, the AI twin material is where I would want more. Creating a synthetic version of yourself that speaks words you did not say is a genuinely consequential thing to publish, and the documentation covers the mechanics thoroughly and the responsibilities not at all.
There is nothing on disclosure, nothing on what happens if the likeness is misused, nothing on the platform rules that increasingly govern synthetic media. Those are not exotic edge cases. They are the ordinary situation for anybody using this feature as intended. As with most of this category, the material also treats output volume as self evidently good.
The Autopilot workflow exists to produce finished video from a prompt with minimal intervention, and the fifty one articles supporting it are all about operating that pipeline rather than about judging what comes out of it. The company is clearly ambitious, with a partnership on an Indian AI film slate and around fifty two million dollars raised, and none of that ambition has been spent on teaching people to make something good. Two point eight. Broad, current enough support documentation for a fast moving product, marked down for rebrand debris, a paywall in front of learning, and silence on the ethics of synthetic likeness.