Why openCanviz starts from your document, not from a prompt
Most AI video tools ask you to describe what you want. We ask you to paste what you already wrote, because the facts you need are already in it.
By openCanviz • August 12, 2026
6 min read
Every AI video tool asks the same opening question: what do you want a video about? You type a sentence, it invents the rest. That works right up until the moment the video says something that is not true about your system, and then you are editing a video to remove a claim you never made.
openCanviz opens with a different question. Paste the thing you already wrote.
The document already did the hard part
A README, a design doc, a lesson plan, a paper. Someone already decided what matters, in what order, and what the tricky parts are. That ordering is the expensive part of making an explainer, and it is sitting in the file.
What the engine actually extracts
Entities and their relationships, the sequence things happen in, the numbers, and the places where the document itself signals difficulty. Those become the spine of the story.
So the first pass is not writing. It is reading. The engine pulls out the facts and the shape, and only then drafts a story that is grounded in them.
Grounding is what makes the edits small
When a draft is built from your document, the edits you make are small and local. A label is wrong, a scene is too fast, a metaphor does not land. You fix that one thing.
When a draft is invented from a prompt, the edits are structural. The video is about the wrong system, in the wrong order, with numbers you have to check one by one. That is not editing. That is rewriting with extra steps.
We measure this. The metric we care about most is not how good the first draft looks in a demo. It is how much of the draft survives contact with the person who knows the subject.
What you still control
Grounding does not mean the engine is on rails.
- 1
The outline
Before anything animates, you see the chapter list and can reorder, cut, or rewrite it.
- 2
The look
Drawn, wall chart, explainer, all plate. The same facts, told in a different visual language.
- 3
Every scene
Each scene is a real editable document, not a rendered frame. Move a box, change a word, retime a beat.
- 4
The narration
Rewrite a line and the scene retimes around it, or record the line in your own voice.
The part that surprised us
We expected people to paste marketing copy. They paste engineering documents. Design docs, incident write-ups, onboarding guides, internal wikis.
Those documents share a property that makes them good raw material: they were written to be correct rather than persuasive. There is a lot of structure in them and very little fluff. The engine has something to hold onto.
If you want to see what that looks like on a document you did not write, the showcase pairs each video with the input it came from.