Prompt Packs
From the Lab

The AI workflows we actually run on client work — packaged, tested, and written for the models we use every day.

There is no shortage of prompt lists on the internet. Most of them are a hundred clever one-liners that work once, in a demo, on the model they were written for, and never quite reproduce when you need them to.

That's not a prompt problem. It's a systems problem.

What we package instead

Everything we release comes out of paid client work. If a workflow hasn't survived a real deadline with a real client on the other end, it doesn't go in a pack. What that gets you is less magic and more repeatability — the same input structure producing the same class of output, run after run.

A prompt that works once is a trick. A prompt that works every time is a tool.

Each pack documents the model it was built for, what it's meant to produce, the failure modes we hit, and how to recover when the output drifts. That last part is the piece nobody publishes, and it's the piece that saves you an afternoon.

What's in rotation

Our working stacks cover image direction, video generation, brand and identity exploration, and the automation glue that ties them together. We build across the tools we run in production rather than committing to a single vendor, because the leaderboard changes every few months and your workflow shouldn't have to.

Drops go to the list before they go anywhere else. If you're after something specific, tell us what you're trying to make — that's usually a faster route than waiting for a pack that matches.

Get the drops first

New packs go to the list before they go public. Tell us what you're building.

Request a drop
← All Lab Notes