The hardest part of running performance ads at a D2C brand isn't finding a winning ad. It's replicating it across markets while getting the culture and feel right at scale. So we built a workflow on Agent One that lets your team localize a winning ad on the same day you identified it. We took three different winning Grüns ads and localized each one for Japan and Spain at ~$70 per ad.

Moving to the video gens – one thing is worth flagging. The reference ad had captions burnt into it. If the agent attached that reference while prompting, the AI video model would pick up those captions and put them into the new gens too. Which I didn't want, because I'd be adding fresh translated captions in post. So I asked the agent to recreate the ad shot by shot using just the character sheet and the location sheet, plus the flow it had broken down earlier. Alternatively, I could've asked the agent to strip the captions off the reference first and then attach the clean version. I then iterated on the shots and locked the Japanese version. I ran the same workflow for the Spanish one, and both ads were ready to ship.

If you’re creating a micro drama series here’s the best way to co-create with your entire team. You're always going to have multiple people working on the same micro drama series. Here's the workflow that keeps everyone aligned. And we're also covering what happens when the character changes mid-series, and how that information passes around the entire creative team automatically.

Penumbra - Episode 2 releasing soon. 2 weeks ago, we launched Epi 1. And Epi 2 is already right around the corner. This just proves that if you have a story worth telling, you don't need to wait for someone in the room to take a bet on you. Quoting the director of this series, Vinu, "Agent One is the closest thing I've found to ten extra hands - I describe the intent, delegate, and it comes back with the shot while I'm already thinking about the next one. The pace honestly doesn't make sense to me yet." Stay tuned.

Here's how to set up Agent One so that you and your team can ship 10 microdrama episodes, in under 3 days. Now, every microdrama company already has pre-production documents — show bibles, scripts, character breakdowns. The best way to use them is to feed them directly to the agent. Doing this means you don't have to waste days and tonnes of credits on repeating yourself every time you're generating shots and running multiple iterations on each.

Jewelry is the hardest category to pull off with AI. The products are intricate, and using a raw model means your product drifts shot to shot. So we picked Tiffany as the example – one of the hardest brands in the world to do this for – and built the three ads that matter most for a brand like this. A brand film, a product film, and an anthem film. Every ad held 100% product consistency. ~$2,400 total. Full breakdown in the video and the thread.

The most powerful moment of the whole project – the coverage sheet. Once the cast and location were locked, and the narrative beats were set, I needed a lot of angles fast. I asked the agent to generate a coverage sheet. A 3x3 grid of nine shots, covering every angle I needed for one narrative beat. Locked the grid and asked it to animate the entire coverage sheet as one continuous video. I took that single video into Premiere and split it into nine individual shots. Did this three times across the setups – three generations, twenty-plus usable shots and the whole ad was covered.