AI can generate a lot. It can’t decide what actually works.

This is my Human by Design entry which is a full product promo teaser, built entirely with envato’s AI toolkit, split into 3 parts in one video:

The hook: setting up the problem before you even see the product.

The build; every prompt, edit, regenerate, and dead end (yes, including the ones that didn’t work)

The result is a final cut polished enough for a real client to post today.

The moment that mattered most: mid-shoot, the AI video generation got flagged and blocked outright. No amount of re-prompting fixed it. So I stepped in and redirected the entire visual direction myself the shots, the pacing, the story beat it was supposed to hit.

AI can execute. It can’t make that call. That’s the difference this challenge is really asking about not what AI can produce, but where a human still has to step in and shape it.

Made with Envato’s Gen AI + stock tools.

Soar for the laurels

You're divinely backed up

Congratulations 🔥🔥🔥

Portfolio update day 4/30:

Today I started migrating the scene to Three.js with Claude. What I've accomplished: WebGPU scene with soft shadows and reflections, transfer of models, materials, and environments, lighting setup. As expected, I transferred all the routine work to Claude and didn't bother with manual export. This will require review and possible optimization in the final stage, but for now the result is amazing. Next up are the bot animation, the Rive interface, and finally, the creation of the website layout into which the scene will be integrated. I keep building!

WebGPU still has spotty browser support so that review stage probably needs a fallback path not just optimization. worth locking in early since it is easy to design around it later than to patch it after the scene is built