Test ten angles, not one guess.
Feed the pipeline a list of hooks and angles, loop over them, and let per-step caching do the math: ten variants cost a rerun, not a production cycle, and every one stays recognizably on brand.
Ten variants per rerun
The hook list is an input. The pipeline loops over it, and only the steps a new hook touches recompute; everything else stays cached.
loop over hooks · changed steps recomputeOn brand at any volume
The brand kit's voice and look steer every script and caption, so variant thirty is as recognizably yours as variant one.
your-brand.kit → every variantShip the winner as an app
When a variant wins, publish its pipeline as a simple form. The team types a hook and presses run, no canvas required.
pipeline → App Mode formBuilt once. Run forever.
Write the list, not the ads.
Ten hooks, ten angles, one text node. The pipeline loops over the list and builds a variant for each entry.
Add an eleventh next week and rerun; only that variant's steps compute, the other ten stay cached.
Every angle, same voice.
The brand kit steers each script, then a video model turns it into the spot. Swap veo-3 for kling-3 without rewiring the graph.
The canvas shows the estimated credit cost of the full loop before you press run, so a big test is never a surprise.
Promote the winner, keep the pipeline.
When results come back, the winning variant's pipeline is already built. Publish it to the team as an app, or trigger batches through the public API.
Pause any step for human review, so nothing ships that a person didn't sign off on.
Every variant, one rerun.
Open a template, paste your URL for the brand kit, and loop your first hook list on the free tier. Costs show on the canvas before every run.
How this works in practice.
What is creative testing at scale?
Running many ad variants, each with a different hook, angle or opening, against each other to find what converts instead of betting on one concept. Orisu makes it a loop: the pipeline iterates over your hook list and builds a variant per entry, with per-step caching so only changed steps recompute.
How do I test different hooks without redoing the whole ad?
The hook is an input, not part of the build. Change the list and rerun: steps that didn't change are cached, so a new hook only recomputes the script, video and captions it actually touches.
How do AI ad variants stay on brand?
Paste your site URL once and Orisu builds a brand kit: colors, fonts, voice and logo. Every script and caption node reads from it, so thirty variants share one personality instead of drifting into thirty.
Do I have to build this pipeline myself?
No. Start from a template, or describe what you want and Orisu drafts the pipeline: nodes, wires and all. You can also build it step by step if you prefer.
What does it cost to run?
Every node shows its estimated credit cost on the canvas, and a run shows its total before you press it. The free tier needs no credit card.
Can my team use it without learning the canvas?
Yes. App Mode turns the pipeline into a simple form: inputs on top, a run button, outputs below. No canvas knowledge needed.
Can I change models later?
Yes. Models sit behind the same ports, so you can swap one without rewiring the workflow.
Test more, guess less.
Start free: paste your URL for the brand kit, load a hook list, and run your first test loop today.