Though it’s easy to become numb to seeing AI videos, generating them yourself can seem a bit magical. You enter a short prompt and boom! A minute later you get a video of aliens playing basketball on Venus.

Current video models can make clips that even a few years ago would have cost thousands to film. They can produce animations that would have required years of training and a lot of skill and patience from a human animator.

This apparent easy firehose supply of any video one can imagine creates a potential trap for business owners. They’ll see an AI ad and think, “Hey that looks good!” Then they’ll use Veo or Sora for a few minutes and be like, “Look at this talking gorilla I’ve made. It only took me thirty seconds!”

But the ad isn’t finished. It’s just an 8-second clip or a collection of clips. So they’ll call in an employee and say, “Here’s an easy job for you. Just finish this AI ad. I’ve already done the hard part. Finishing it should only take you an hour.”

The employee goes away, then eventually comes back, with an ad nobody is happy with. It took way longer than expected, doesn’t match the business owner’s vision, and it wasn’t easy for the worker to make after all.

What went wrong?

To understand this, it’s good to lay out the parts that go into a finished video. These are:

1. The footage. This can come from AI generations, or it can be filmed on a camera, or it can be created by an animator.

2. The audio. This can also come from AI or be produced traditionally.

3. Editing. Currently this is still a human task. You can’t go to Veo and say, “Put these five clips together, use this audio, trim this, adjust that.” We’ll probably get there some day, but not today.

4. Scripting, inspiration, matters of taste. Though ChatGPT can write prompts that the video AI uses, ultimately some person has to give direction for what should be in the video, regardless of how the prompts are created.

It’s easy for people to focus on how quickly AI can spit out a short clip, and dismiss all the human factors that still go into making a proper video out of what the AI has generated.

On top of this, as anyone who has used Veo or another video model knows, using AI is no guarantee of getting the clip you want. There’s still a lot of troubleshooting and experimenting involved, particularly if you want something specific or with character continuity. (I wrote more about one common frustration here.)

To make a good AI ad still requires a lot of work, skill, and judgement calls by a person. Yes, it cut out the need for the film crew or animator, but that’s only one part of a video.