07/21/2025
In my continued exploration of AI-generated video, here's a fanciful TV ad demo. I've achieved perfect character consistency, which has been a challenge in the first month of my experimentation. AI produced all the imagery with a little help from me and Photoshop. I use the generated video clips as my raw footage. That's then edited (I keyed elements in some scenes, others have rotoscoped elements), graded, titled, and captioned. Sound design elements are created in Pro Tools using Krotos tools, such as DeHumanizer, for the spider roars and screams. As a director, I have complete control over every element of an AI video, provided the AI model permits it. The ongoing challenge is to create prompts that do not cause AI hallucinations. The best prompt is up to 600 characters, which is a challenge. I'm learning the details of how AI models interpret text descriptions of visuals. It doesn't "think" in visuals. It requires specific instructions. The models are improving, so we can now upload guide images that the AI uses to maintain consistency. This means we can include accurate product imagery, if needed. Using AI-generated video remains both frustrating and exhilarating. But, I'm getting there.
