To test Piveth and Astra versions of the same SKU ad, generate both variants from the same photos in the same order, review each for garment fidelity, then run them simultaneously in your ad platform's native split-test tool against your current best ad as a control, with one primary metric and stopping rule set beforehand.
Your best-selling dress ad is fatiguing, and there is no budget to reshoot it. Refreshing creative usually means booking a studio, a model, and a stylist, which is hard to justify for a single SKU.
Take a Shopify seller with only a flat lay and a ghost mannequin shot of one dress. She wants two fresh video directions without a new shoot. With FashionUGC, she can turn clothing photos into video and produce a Piveth version and an Astra version from those same two images. Piveth and Astra are the two generation variants named in FashionUGC's product docs.
Set expectations first. FashionUGC creates the video files, and a person reviews them. The tool does not publish ads or promise performance.
This guide covers the requirements for a fair test and six steps to run one. It also covers the one step most teams skip, which can invalidate the whole test before it launches.
What Does a Fair Same-SKU Creative Test Need?
A fair same-SKU creative test changes only one variable, which here is the Piveth or Astra variant. Everything else stays locked: the product, the source photos, their order, the copy, the audience, the placement, and the timing. That way, any difference in results points to the video itself.
A SKU-level test compares ads for one specific product, such as one dress in one colorway. A control is your current best-performing ad, which serves as the baseline. A single variable is the one element you allow to differ.
Think of it like a taste test where both cups hold the same coffee and only the brewing method changes. If you swap the beans too, you can't tell which change mattered.
Prerequisites
- 1 to 6 sharp source images of one SKU
- Visible hems and clean, even lighting
- 9:16 vertical output for Reels, TikTok, and Shorts
- An existing best-performing ad to use as the control
- Access to the native A/B or split-test tool in Meta or TikTok
Should both variants use identical source photos? Yes. Generate Piveth and Astra from the same images, uploaded in the same order. The photo set is the input layer, and generic A/B testing advice usually skips it. If one variant gets an extra on-body shot or a reordered sequence, you are testing two variables at once. A lift in click-through rate could then come from the better input image rather than the variant, and you can't separate the two effects afterward. FashionUGC accepts between one and six photos per item, so upload the set once and reuse it for both generations.
| Element | Locked or Variable | Why |
|---|---|---|
| SKU | Locked | Different products sell differently. |
| Photo set | Locked | Input quality shapes the output. |
| Image order | Locked | The sequence can change which view leads. |
| Caption or copy | Locked | Copy changes click intent. |
| Audience | Locked | Segments respond differently. |
| Placement | Locked | Feed and Stories behave differently. |
| Budget split | Locked | Uneven spend skews delivery. |
| Schedule | Locked | Timing affects auction costs. |
| Video variant (Piveth or Astra) | Variable | This is the only thing you are testing. |
Before you generate anything, read the guide on how to prepare product photos for AI fashion video.

How to Test Piveth and Astra Versions Step by Step
Prepare one photo set, generate both variants from it, review the videos for garment fidelity, and then run them alongside your control in a native split test. Before you start, you need your SKU photos, a FashionUGC account, your current best ad, and access to Meta or TikTok split testing.
- Prepare one photo set. Select 1 to 6 images of the SKU: sharp garment plates with visible hems and no props. A flat lay and a ghost mannequin shot are both valid inputs, as covered in the guide on how to turn a flat lay into a fashion video. Time: about 10â15 minutes. Success looks like one folder of final images in a fixed order.
- Generate both variants. Upload the identical set in the same order, then choose 9:16 output. Generate Piveth first, then repeat the process with the same inputs for Astra. Success looks like two downloadable video files built from the same source.
Here's the catch: a variant can win the test while misrepresenting the garment, and that can lead to returns.
- Review both videos before testing. This is the step most teams skip. Watch each file at full screen and check the following:
- The colorway matches the source.
- Hems are visible and intact.
- Logos and prints are legible.
- Fabric texture is preserved.
- The drape is plausible.
- The fit is not distorted.
- Upload the files and your control. Open the native A/B or split-test tool in Meta or TikTok, then add the Piveth file, the Astra file, and your current best ad. For example, the dress seller loads all three ads into one split test with the same copy and audience. Success looks like three ads in a single test.
- Set the metric and stopping rule before launch. Choose one primary metric, such as hook rate, CTR, or CAC. Write down when you will stop the test, then run all three ads at the same time. Success looks like a live test with documented rules.
- Read the results. Compare each variant against the control using the criteria in the next section.
Why review AI fashion videos before running a split test? A split test measures attention, not accuracy. If the Astra version shifts a navy dress toward black, or the Piveth version shortens a hemline, the ad may earn clicks while showing a product the customer will not receive. That gap can turn a winning CTR into returns and lost trust. Reviewing colorway, hems, prints, texture, drape, and fit takes a few minutes per file. It also ensures the test compares two honest representations of the same SKU, so whichever variant wins does so fairly.
Ready to refresh a fatigued ad? Use the AI clothing ad generator to create both variants from the photos you already have.
Once your test has run, here is how to read the results.

How to Tell If Your Piveth vs Astra Test Worked
Your test produced a valid result only if the metric and stopping rule you set before launch were met while every other element stayed locked. A valid same-SKU result is a performance difference you can attribute to the Piveth or Astra version alone, not to budget, timing, or copy.
But wait, before you scale the winner: check for broken locks. A stopping rule is the condition you set in advance for ending the test, such as a spend level or a date. Before trusting any result, confirm both variants ran on matching schedules, matching budgets, and identical captions.
Think of it like a blind taste test. If one cup is served colder, you are no longer judging the coffee. Run your raw numbers through the ad metrics calculator so both variants are compared using the same formulas.
Then expect a surprise. The variant you preferred in review may lose, and neither may beat your control, which is the existing ad you are trying to replace. Treat TikTok and Reels as separate tests, because a winner on one platform says nothing reliable about the other.
When a variant looks off, trace the issue back to the photo inputs:
| Symptom | Likely cause | Fix |
|---|---|---|
| Drape looks wrong | Input was a flat lay | Add a ghost mannequin or detail angle |
| Colorway differs between variants | Soft or poorly lit front plate | Regenerate from a sharper front plate |
| A detail is lost | Low-resolution source | Supply a higher-resolution still |
Worked example #3: one variant shifts a dress's colorway, so the seller regenerates it from a sharper front plate and relaunches before reading any results.

Piveth and Astra Testing FAQ
These answers cover platform splits, test length, losing variants, and what FashionUGC does and does not handle.
What if my TikTok result doesn't match Reels?
Treat each platform as its own test. In a split test of TikTok vs. Reels creative, audiences, placements, and viewing habits differ, so conflicting winners are normal. Keep the platform-specific winner on each platform rather than forcing one variant across both channels.
How long should I run a test before calling a winner?
Run it until the stopping rule you set before launch is met. Choose your success metric and that rule before the first impression is served. Ending early because one variant looks ahead reintroduces bias, and no fixed duration or budget guarantees a conclusive result.
What if neither variant beats my current ad?
Keep the control running and revisit your photo inputs. A losing pair often points to soft front plates, missing detail angles, or flat lays that misrepresent drape. Regenerate from stronger source images, then retest against the same control.
Does FashionUGC run or publish the test?
No, FashionUGC does not run or publish the test. It creates the video files, and you review them before running the test in your own ad platform. FashionUGC does not publish ads and does not promise performance results.
Run Your First Same-SKU Test
Your first trustworthy test starts with one fatigued SKU and the photos you already have. The finding teased at the start holds: the version you like best in review is not automatically the version that wins.
Two habits make any result worth acting on. The first is the One-Variable Rule, which means only Piveth versus Astra changes. The second is the review gate, which means no file launches until it passes fidelity checks for color, drape, and detail.
- Pick one SKU whose current ad is fatiguing, and gather its existing 1 to 6 photos.
- Generate both versions, then pass each through the fidelity review.
- Launch both against your control, using a metric and stopping rule you set before launch.
Upload one SKU's photos to the AI fashion UGC video generator. FashionUGC creates the files, you review them, and no performance outcome is promised. Each test you run makes your next creative decision a measured one rather than a guess.