fairytale video
Make an ad
Ad creative testing with AI video variants
Two ads for the same product, with the same budget and the same audience, can perform very differently. Most of the difference is the creative: the opening, the person, the setting, the words. Creative testing is how you find out which version works instead of guessing, and AI video makes it far cheaper to run, because each variant takes minutes rather than a new shoot.
Test one thing at a time
If two ads differ in their hook, their creator and their setting, and one wins, you've learned that one ad beat another, not why. Change one thing per test and keep everything else identical. AI video helps here: reuse the same product, character and scene references to hold the inputs steady. Generated motion and delivery can still vary, so inspect every result and avoid treating one pair of videos as proof.
Take these two. They share a format (a creator at a bathroom mirror, one makeup product, a short reaction) but differ in the person, the product, the setting and the line all at once. If one beat the other, you couldn't say why. To learn something, change just one of those.
What to test, in order
- The hook. The first two seconds: the opening action and the first line. It decides whether anyone sees the rest, so it's usually the first thing to test. 30 UGC ad hooks gives you a starting set.
- The format. A first-use ad against an unboxing against a POV. 12 AI video ad ideas has formats to compare.
- The person. The same ad with different creators, to learn who your audience responds to.
- The setting. The kitchen against the car against the gym.
- The line. The same action with different wording, or with no dialogue at all.
- The finish. Caption style, length, the closing shot.
How many variants, and how long
Three to five variants per round is a manageable number. Run them at the same time, to the same audience, in the same placements, with the same budget each, so the creative is the only difference.
Give them long enough. Early numbers jump around; a variant that leads after a few hundred impressions often doesn't lead after a few thousand. Decide in advance how many impressions or how many days you'll wait before calling a winner, and stick to it.
Which numbers to read
- Hook rate: the share of impressions that turn into views of the first few seconds. The best signal for hook tests.
- Hold rate or average watch time: whether people stay after the hook. Low hold with a high hook rate means the opening promised something the rest didn't deliver.
- Click-through rate: whether the ad makes people want to know more.
- Cost per result: cost per purchase, sign-up or whatever the ad is for. This decides the winner; the others explain it.
Each platform names these slightly differently, but all of them report something close to each.
Name everything
A simple naming pattern saves hours later. For example: product_format_hook_creator_scene_v1, such as mug_firstuse_leakproof_maya_car_v1. Anyone looking at the results can see what each variant was without opening it.
Turn a winner into the next round
- Keep the winning hook. Test it with two or three different creators.
- Keep the winning creator. Test the hook in two or three settings.
- Once an AI version has clearly won, consider briefing a real creator to film the same angle, and test the two against each other.
- Keep a log of what won and what lost. Losing variants teach you as much as winners.
Watch for fatigue
Even a winning ad wears out as the same people see it again and again. When its cost per result climbs while nothing else has changed, it's time for the next round. If you're always testing, you always have a replacement ready.
Testing with fairytale video
fairytale video is built for making variants. Reuse the same creator, product and scene, change one thing such as the opening line, and generate another video with Seedance 2.5. Every generated video uses credits, so set a production budget for the round and review your brief before rendering. Each video keeps the concept, timed plan, exact prompt and references it was made from, and later edits to the ad never change an existing video. Keep the original ad when making variants, so you can compare the inputs and results. For the steps behind each one, see how to make an AI ad.