What my first AI video project taught me

Last week I got my first ever project making an AI video.

I just said yes, because I wanted to push my limits. I took on this Candi project with a deadline of about 16 hours. Ha.

I’d never made a full AI video before, so I worked out the workflow as I went, learning along the way. I used two tools: Google Flow and ChatGPT.

It was pretty chaotic, but that’s exactly how a lot of things only really sank in: by doing them hands-on.

The weak spot wasn’t the AI

Visually, I think the result is pretty good for a first project. Character consistency, ambience, camera movement, and even a few of the shots turned out better than I expected.

But after watching the final result again and again, what felt most lacking wasn’t actually the AI generation. It was the storyboard and the transitions between scenes.

One scene can be good on its own. But once it’s joined to the next scene, the story doesn’t necessarily flow smoothly.

What I took from it: making AI video isn’t just about being good at prompting. You still have to know how to tell a story.

What I want to learn next

So what I actually want to dig into next is:

• Storyboarding
• Continuity between scenes
• Camera language
• Transitions
• Visual storytelling

The plan is to collect reference films, tutorials and video breakdowns in NotebookLM, then use ChatGPT to help me pick apart the patterns one by one. I’m also still looking for good channels, courses and resources for learning AI filmmaking and storytelling.

Rather than just learning “good Veo prompts”, I think it’s more interesting to learn to think like a filmmaker at the same time, and then let AI be the tool for execution.