Google Omni, AI Video Creation and the Importance of Prompt Frameworks

I recently started experimenting with Google Omni for AI video generation while working on promotional material for my upcoming children’s book Drawing Hope: The Magic Pencils.

The experience was both exciting… and frustrating.

Like most people experimenting with AI video right now, my first attempt looked incredible for about 80% of the clip. The lighting was cinematic. The movement felt emotional. The atmosphere was almost exactly what I had imagined when writing the story.

And then the hallucinations started.

The AI changed the book cover halfway through the sequence. Text became misspelled. Typography shifted between scenes. Small visual inconsistencies appeared that completely broke immersion. The overall structure was beautiful, but the reliability of the output wasn’t stable enough to trust on its own.

That was the moment I realised something important:

AI video generation is not simply about creativity anymore.
It is about instruction architecture.

The quality of the result is becoming directly tied to the quality of the framework you build around the AI.

The first version of the video failed because my instructions were too broad. I treated the AI like a creative collaborator instead of a system that needed operational boundaries. Once I changed that mindset, the results changed dramatically.

Instead of simply saying:

“Use this book cover in the video”

I started building constraint-based prompting.

I began explicitly instructing the model what it was NOT allowed to change.

Things like:

• preserve composition
• preserve lighting
• preserve typography
• do not redesign the cover
• no spelling modifications
• maintain the exact camera movement
• treat the supplied asset as fixed

The difference was enormous.

The second-generation video suddenly became coherent. The hallucinations reduced dramatically. The typography stabilised. The emotional tone remained consistent throughout the sequence. Most importantly, the book itself remained visually accurate from scene to scene.

What became obvious very quickly is that AI systems naturally “want” to creatively reinterpret material unless you define operational guardrails.

That is where frameworks become essential.

In many ways, this mirrors the same governance discussions happening across enterprise AI right now. Most AI failures are not because the models are unintelligent. They happen because humans provide vague instructions without defining acceptable operational boundaries.

What I was building for a children’s book trailer was actually a miniature AI governance system.

The prompt evolved into something far more structured:

Use the original video as the motion and composition reference.

Only replace the incorrect 3D book cover with the supplied

3D book cover image.

Do not change anything else in the video.

Keep:
- the same camera movement
- the same lighting
- the same background
- the same timing
- the same particles
- the same music mood
- the same ending
- the same overall scene composition

The supplied 3D book cover must be used exactly as provided.
Do not redesign it.
Do not recreate the cover art.
Do not change the title, typography, colours, spine,

page angle or book shape.

Treat the supplied 3D book cover as a fixed printed physical 

object placed into the existing scene.

Remove the incorrect book cover currently shown in the video 

and substitute this exact supplied 3D book cover in the same

position, scale, angle and movement.

Critical:
No text changes.
No new cover design.
No typography hallucination.
No spelling changes.
No changes to Hope’s face.
No changes to the background or other elements.

Only change the 3D book cover.
Everything else stays the same.

What fascinated me most was how quickly the AI improved once the ambiguity was removed.

This is where I think a lot of AI development is heading.

The future will not belong only to people who can “use AI.”
It will belong to people who can structure reasoning, constraints and instruction systems around AI.

Prompting is evolving into operational design.

And honestly… watching a cinematic trailer emerge from a story I wrote while living on Russell Island was one of the strangest and most exciting creative experiences I’ve had in years.

Not because the AI replaced creativity.

But because it amplified it — once the framework became strong enough to guide it.