Let's Heartie

What if your AI agent was a 5-foot, fluffy monster? This is "Let's Heartie"—a "CG + Live-Action" AIGC IP advertisement where connection makes reality bloom.

Comment

Inspiration: The "Emotional Ecosystem"

This project is my personal "dragon", a challenge born from a simple thesis: "If you can make socializing feel safe for someone with Asperger's, you can make it feel safe for everyone."

The inspiration for "Let's Heartie" wasn't a product, but an emotional ecosystem. Heartie is our original IP, an AI agent designed for "safe socializing."

Our visual inspiration was Paddington Bear. The goal was a 90-second "world's breath," using the AIGC challenge of "CG + Live-Action" to bring a human-sized, fully realized Heartie into real-world Sydney, seamlessly blending magic with reality.

How It Was Built (The Workflow)

This ad is a 90-second visual symphony, driven entirely by the music's rhythm, not a linear script. It’s built on a foundation of hard cuts, high-speed rhythm, and relentless IP iteration.

My workflow was a directorial hybrid:

  1. IP Lock-In: We first secured a single, 100% perfect holy grail generation of our Heartie IP. This safe asset became our anchor.
  2. CG+Live-Action Fusion: We used this holy grail IP as the primary Ingredient in VEO 3.1 to generate all subsequent live-action shots in Sydney, forcing the AI to maintain consistency while building a photorealistic world around it.
  3. Rhythm-Based Editing: The entire film was cut directly to the music's sections (Intro, Verse, Chorus). This meant abandoning seamless transitions in favor of the absolute control that hard cuts provide over the emotional pacing.

Challenges

This project was a 48-hour war against VEO 3.1's hallucinations.

My single greatest challenge was IP Integrity. VEO 3.1 constantly tried to "betray" the Heartie design. It gave it ugly tails, long legs when we asked for human-scale, pink wings, or crystal balls when we asked for lens fog. I think the reason is i was using ingredients to video instead of frames to video, which increases uncertainty.

What I Learned (My "Dragon"):

  1. VEO 3.1 is a Style Generator, Not a Physics" Engine: It cannot handle complex physical interactions (like nuzzling) or compound instructions (like 3 IP characters + selfie logic). It will hallucinate and break your IP.
  2. Hard Cuts Are Not Weakness; They Are Control: Relying on AI for transitions is soft. Using a hard cut is a director's decision that gives you 100% control over rhythm, pacing, and IP safety.
  3. You Must Be "Un-f*ck-with-able": When the AI fails, you do not compromise your vision (like accepting an ugly tail). You find a way by changing the tactic (the Prompt), never the goal (the IP).

This film is the result of that "un-f*ck-with-able" persistence.

Built With

  • english
Share this project:

Updates