What happens when you take a hero frame from an AI-generated video and push it back into an image edit model?

This weekend I took a still from a recent Seedance 2.5 piece I made —depicting a solo mountaineer in a harsh alpine environment — and ran it through Midjourney's new v8.2 edit model to see how far I could iterate on a single frame without regenerating from scratch.

Tested wardrobe swaps, environment shifts, and style transfers.

Video walkthrough of the process coming in a separate post.

This looks cool

"Zahid has successfully delivered and deployed HCM within our company."

That's from the client review that landed on my profile this week, for OneHub — the payroll and people-operations platform I built for TechFoundry. Monthly payroll with fiscal-year tax withholding, payslips and tax certificates on demand, expense reports read line by line by Claude, and one reporting layer so the dashboard, the payslip and the certificate always agree.

The word I care about in that review is "deployed". Delivered means the code works. Deployed means payroll runs on it every month, people are paid the right amount, and nobody has to call me.

The thing I'd do differently: the tax rules started out hard-coded. Fine for one country, one fiscal year — until it wasn't. We later pulled them out into a rules engine so the same payroll runs across countries without touching the code. Next time that's day one, not a refactor.

It's also a Contra-verified project — the work was paid through Contra, so the review and the case study are tied to the same engagement. Case study: https://contra.com/p/Y3EgrEUc-payroll-and-people-operations-platform

Shipping a complete payroll and fiscal-year tax withholding engine straight to production is a major milestone! Full-stack execution at its finest.