Black Friday is won by the team with fresh creative to rotate when costs climb and October's best ad stops converting. Production used to be the limit: a studio can't turn around 40 new variants in a week. With generation, the limit is planning. This four-week sprint takes you from last quarter's winners to a tested BFCM library with backups approved before the sale. Black Friday 2026 falls on 27 November, so four weeks out is the end of October. On 8frame a 60-clip library costs about 4,480 credits on Kling v3, or about 3,100 if you test on a budget model and redo only the winners on Kling.
Correction, 2026-09-30: the first version of this page described a batch node with prompt injection and parallel jobs (you build one node per variant on the canvas), priced Kling v3 clips at $0.34 (57 credits for 5 silent seconds) and Nano Banana Pro at $0.04 to $0.08 (21 credits), described a sample brand's sprint with costs and results we can't reproduce, and quoted ad review times as verified against platform guidance when we couldn't confirm them. This version keeps the plan and uses registry prices.
TL;DR
- Week 4 (four weeks out): audit what converted, build a hook bank, lock your product references. No generation.
- Week 3: generate the library: 40 to 60 clips across your top 3 to 5 offers. Kling v3 is 57 credits for 5 silent seconds.
- Week 2: test at low spend, find the 8 to 12 hooks worth scaling, freeze the account structure.
- Week 1: scale winners, stage backups, and get everything through ad review early, because review queues get longer at peak.
Why creative volume is the BFCM lever
During the sale window your audiences are the same, your budgets are elevated, and everyone in your category is bidding on the same people. The one variable you still control is creative freshness. Creative fatigue accelerates under high frequency: an ad that held for three weeks in October can burn out in 48 hours at Black Friday spend levels because the same user sees it five times a day instead of once.
That is why the sprint front-loads generation. You want a reservoir of tested variants sitting ready so that when a winner fatigues on the Saturday of the sale, you swap in the next-best hook the same hour instead of briefing a shoot you won't get back for two weeks. For the mechanics of finding winners from a batch, the AI ad variant testing workflow is the engine this sprint runs on.
Week 4: Winners audit and hook bank
No generation this week. You are deciding what to make.
Pull your last two quarters of ad data. Sort by hook retention and cost per purchase, not by spend. You are looking for the hooks and angles that already worked, because those are your safest bets to rebuild at volume for a high-stakes window. Write down the top 8 to 10 performing hooks in plain language: "before-and-after in the first two seconds," "founder talking to camera about the discount," "unboxing with the price on screen."
Build the hook bank. Take those proven hooks and add 6 to 8 new ones tuned for BFCM specifically: urgency framings ("lowest price we run all year"), gifting angles ("the one everyone actually keeps"), and bundle offers. You now have a bank of roughly 15 hooks. That is your test surface.
Lock the concept references. For each product going into the sale, capture one clean reference image: product on white, plus a lifestyle reference if you have one. These become the locked inputs for every clip so the batch stays on-brand. If you're generating product stills, Nano Banana Pro makes them from your product photo at 21 credits each.
Week 3: Batch generation
Generate the whole library in one or two sessions. On the canvas: your locked product reference as an image node connected to one Kling v3 node per variant, each with one hook from the bank as its prompt; duplicate a node and swap the hook to make the next. Kling v3 follows the start image's shape, so prepare a 9:16 and a 1:1 version of each product reference for the two passes.
For a store with 4 hero products:
| Line item | Quantity | Credits |
|---|---|---|
| Kling v3 clips, 5 s silent, 9:16 | 40 | 2,280 |
| Kling v3 clips, 5 s silent, 1:1 | 20 | 1,140 |
| Nano Banana Pro reference stills | 8 | 168 |
| Veo 3.1 hero clips, 8 s silent | 4 | 896 |
| Total | 72 assets | 4,484, about $45 |
A budget route: run the 60 test clips on Hailuo H3 Max Turbo at 768p (27 credits for 5 seconds, 1,620 for 60) or Wan 3.0 at 480p (24, 1,440 for 60), then regenerate only the 10 or so winners on Kling v3 (570). With the same stills and Veo heroes, that's 3,074 to 3,254 in total, against 4,484.
Reserve Veo 3.1 (224 credits for 8 silent seconds, 448 with audio) for the two to four hero placements where light, cloth and depth carry the ad.
Review every clip before it enters the library. Colour temperature and product sheen wander across a large batch; flag outliers and re-run them. A BFCM ad with the wrong product colour is worse than no ad.
Week 2: Variant testing
You have a reservoir. Now you find the winners before the sale, at low spend, while impressions are cheap.
Group variants by hook, not by product, and launch 10 to 15 of them at a modest daily budget across a few ad sets. After 72 hours, cut the bottom half on hook retention and note the 8 to 12 hooks that separated from the field. Those are what you scale during the sale.
Freeze the account by the end of this week. Meta's learning phase needs lead time, and every structural edit during the sale resets it. Lock your campaign structure, audiences, and budgets now so that in Week 1 you are only swapping creative, never rebuilding ad sets. Getting disclosure right is part of the freeze too: check each platform's current rules on AI-generated content; the rules on disclosing AI-generated ads are worth a read before you scale.
Week 1: Scale winners and load backups
Scale the proven hooks into your frozen structure and raise budgets on the ad sets carrying them. Keep 3 to 4 winning variants live per ad set so delivery has room to rotate as frequency climbs.
Load backup creative for fatigue. Stage your next 10 to 15 tested-but-not-live variants. When a winner's cost per purchase climbs mid-sale, swap the next hook in the same hour.
Get through review early. BFCM is the busiest advertising window of the year and review can take longer than usual. Submit everything, including backups you may never run, several days before you need it, and launch sale ad sets early enough to exit the learning phase before budgets scale. Check Meta's and TikTok's current guidance on review times; an approved ad sitting paused is worth far more on the Saturday of the sale than a great clip stuck in a queue.
Pitfalls
Generating in Week 1
If your library isn't built until the sale is live, you have no reservoir and no backups. Generation is cheap and fast, but variant testing needs 72 hours to produce signal. Front-load generation into Week 3 so testing finishes before the sale, not during it.
Editing the account mid-sale
Every audience change, budget restructure, or new ad set during BFCM can reset the learning phase at the worst possible moment. Freeze in Week 2 and treat creative swaps as the only permitted live change.
Ignoring the review buffer
The most expensive mistake of the sprint is a winning backup that can't go live because it entered a long review queue on Black Friday morning. Push everything through review early, including ads you might not run.
FAQ
How many ad variants do I need for Black Friday?
Plan to test 40 to 60 and run far fewer live: 3 to 4 per ad set across your winning hooks, with 10 to 15 tested backups staged for fatigue swaps.
When should I submit ads to avoid the BFCM review queue?
Days early, not hours: review can take longer at peak. Get backups approved early too, so they're ready to unpause when a winner fatigues.
What does the creative library cost?
About 4,484 credits for 60 Kling v3 clips, 8 Nano Banana Pro stills and 4 Veo 3.1 heroes on 8frame, or 3,074 if you test on Wan 3.0 at 480p and regenerate only the 10 winners on Kling.
Can I reuse this sprint for other sale events?
Yes. The four-week structure works for any high-stakes window: back-to-school, Prime-style events, an anniversary sale. Duplicate the board, swap the product references and hook bank, and re-run.
Sources
- 8frame Kling, Hailuo, Wan, Nano Banana and Veo prices: the 8frame tool registry, 2026-09-30
Build the reservoir four weeks out and get it approved before the queue backs up. For shot structures, see the UGC ad template and the product video template. The 8frame canvas is free and unlimited; generation is paid from $19/month.