AI video ad packs are batches of video ad variants built from one brief and split across hooks, formats and audiences. A common grid is three hooks by three formats by two audiences. That grid does not say what the eighteen clips cost. In DesignerBox the cost of one clip changes more than tenfold between models, so pick the model and the pack size together.
You have seen the grid. Three hook families, three formats, two localizations, eighteen assets, ship it Monday. It is a tidy number, and for static creative it is at least arguable. For video it needs pricing before anyone commits to it. The grid also says nothing about how many cells your ad account can read.
Static variants are close to free once the brief is set, which is why the static conversation is about whether your ad account can read the variants you made. Video adds a second constraint on top of that one. Video cost grows with every second of output, and the model you pick moves it the most.
This covers how to price a video ad pack, which models can technically produce the shapes a pack needs, what to vary between variants, and how to size the batch so you do not pay to generate creative you cannot run.
Key Takeaways
- An 18-clip pack multiplies the cost of one clip by 18, and every retake adds to it. The model you pick sets the cost of that one clip.
- Model choice and pack size are one decision. Picking the model first and the count second is how teams commit to a batch they cannot afford to finish.
- Meta recommends no more than five ads per ad set (facebook.com, September 2026). The “six or fewer” line was on Meta’s ad volume page in February 2025 and is not on the live page today.
- Meta does publish a learning threshold: about 50 results per ad set in the week after the last significant edit (Meta Business Help Center, October 2026). That governs how many cells you can read.
- TikTok publishes a number and it is small. Three to five unique creatives per ad group, three to five ad groups per campaign (ads.tiktok.com, October 2026).
- Kling 2.6 makes silent video at 720p. Sound needs 1080p Professional mode, so check the cost before the run on a sound-on Kling pack.
- Roughly half of ad creatives are switched off before 28 days and only 4% to 8% become winners, across 578,750 creatives and $1.29 billion in Meta spend (Motion, October 2026).
What is an AI video ad pack?
An AI video ad pack is one batch of video ad variants generated together from a single brief, then launched as a structured test rather than as individual ads. A pack normally varies three things: the hook, the format or aspect ratio, and the audience or message. Batching works because the expensive thinking happens once. The brief, the product photo and the brand rules are set before the first render, so each additional variant costs only its generation time and its credits. What the pack does not include is the assembly: every clip in it still has to be cut, captioned and exported before it runs.
The pack is a production unit. It is not a delivery unit. Platforms deliver at the ad set level. So the pack has to be split across ad sets before anything runs, and each of those ad sets needs its own budget.
Why the three by three by two grid breaks on video
The grid answers a production question. Three hooks are easy to write, three formats cover the placements, two audiences double the count without doubling the thinking. Eighteen assets feels substantial and takes an afternoon. Once the pack runs, hook rate shows which of the three hooks stopped the most people.
It is silent on the two questions that decide whether the pack ships. Where do the eighteen ads go, and what do they cost to make.
The first question is a media planning problem, and it is the same problem whether the creative is a still or a clip. It is covered properly in how many ad variants your account can read, which works through the division from Meta’s published learning threshold. The short version is that your weekly conversion volume, divided by about 50, gives the number of ad sets that can stabilize, and that is your cell count.
The second question is where video separates from static.
What an AI video ad pack costs
Pack size multiplies against the cost of one clip on the model you pick. In DesignerBox, an 8-second clip costs 40 to 560 credits, depending on the model. The cost is shown before the run. Count the clips first:
| Pack size | Clips to run, one take each | Clips to run, two takes each |
|---|---|---|
| 6-clip pack | 6 | 12 |
| 12-clip pack | 12 | 24 |
| 18-clip pack | 18 | 36 |
Then price one clip on the model you plan to use, and multiply. The same 18 clips cost very different amounts on a light model and on a premium one. No other decision in the pack moves the number that much, which is why model choice is more than a quality preference to settle later. It sets the maximum pack size you can build.
Uploading your own photos and the commercial license start on the Pro plan. AI video and the video editor start on the Premium plan. The free plan cannot make video. Plans and credits are on the pricing page. A still variant costs less to make than a clip, so packs that carry still variants alongside clips stretch a budget further. Image ads and when one image beats a video covers the specs for the still half. The wider breakdown is in what AI video costs.
The practical sequence is to fix the cell count from your media plan first, then pick the cheapest model that carries the look, then check the multiplication before running anything. The table counts one take per clip in its first column, so budget for the takes you discard on the way to one shipped ad on top of it.
Which models can produce the shapes a pack needs
Pack composition hits technical limits before it hits creative ones. Clip length and vertical output are the two that matter most. Every spec below was checked against the provider’s own documentation in September 2026.
| Model | Clip length | Native audio | Vertical 9:16 |
|---|---|---|---|
| Veo 3.1 and Veo 3.1 Fast | 4, 6 or 8 seconds | Always on | Yes |
| Seedance 2.0 | 4 to 15 seconds | Optional flag | Yes, to 4K |
| Kling 2.6 Pro | 5 or 10 seconds only | Yes, 1080p only | Yes on text-to-video |
| Runway Gen-4.5 | 2 to 10 seconds | Not documented | Yes, 720x1280 |
Four constraints matter specifically for batch work.
Kling 2.6 makes silent video at 720p. Sound needs 1080p Professional mode, and audio is off by default, so check the cost of a sound-on Kling pack before the run (kling.ai, September 2026).
Kling’s image-to-video endpoint has no aspect ratio parameter. Output inherits the ratio of the input image, so a vertical pack built from a product photo needs a 9:16 source before generation starts.
Veo 3.1 requires 8-second clips for 1080p or 4K output (ai.google.dev, September 2026). A pack of 6-second Veo clips is a 720p pack, and longer clips cost more.
Runway’s current Gen-4.5 spec and API list no audio output (docs.dev.runwayml.com, October 2026), although Runway announced native audio for Gen-4.5 in December 2025 (runway.com, October 2026). Treat Gen-4.5 as silent unless you confirm otherwise in the product. Runway alternatives sorted by job covers the other options for product and ad work. Once shape has ruled models in or out, the video models side by side helps you pick between the rest.
What to vary, and what to hold fixed
Meta’s December 2025 guidance on creative diversification prescribes no number of variants. It describes a kind of difference instead: assets that are “truly different in look, feel, storyline, and message,” tailored to different customer personas or use cases. Meta distinguishes this from creative iteration and manual refinement (Meta for Business, December 2025).
That is the useful instruction, and it is the one a grid quietly ignores. A pack of eighteen clips differing only in their first two seconds is one creative tested eighteen times, and the platform will read it as one creative.
Hold fixed across the pack: the product, the brand palette and type, the logo treatment, the aspect ratio within a given ad set, and the offer. These make the pack recognizably yours and make the results comparable.
Vary across the pack: the opening three seconds, the narrative structure, the setting, the presence or absence of a person, the pacing, and the call to action. Vary one axis at a time within an ad set so the comparison means something, and change axes between ad sets.
Holding the fixed half fixed is a workflow problem more than a prompting problem. When every asset comes from the same product photo and the same brand profile, consistency stops depending on anyone remembering to check it. The same product, palette and logo appear in every variant. That mechanic is covered in keeping brand consistency across AI creative.
Where the numbers in this category come from
Some pack-sizing figures have a primary source and some do not. The table shows the status of each one. Check it before you use a figure in your plan.
| Commonly repeated | Status |
|---|---|
| ”Meta recommends six or fewer ads per ad set” | Was on Meta’s ad volume page in February 2025. Not on the live page today |
| ”Keep frequency under 2.5, or 3, or 3.5” | No primary source. Meta’s frequency page defines the metric and names no threshold |
| ”Refresh creative every 7 days” | Not TikTok’s guidance. Its creative best practices page ties refresh to declining delivery. Its video advertising guide says every 2 to 4 weeks, or sooner if click-through or view-through rate declines |
| ”50 conversions per variant before you read it” | No Meta or Google documentation states this |
| ”86% of buyers use generative AI for video creative” | Real, but it is IAB’s 2025 figure and it combines using with planning to use (iab.com, July 2025) |
The last one comes from IAB’s 2025 Digital Video Ad Spend and Strategy report, and the correct phrasing is “using or planning to use.” IAB’s newer work is better documented. Its January 2026 study found 83% of ad executives say their company has deployed AI in the creative process, up from 60% in 2024. The study surveyed 104 US ad executives at companies spending $1M to $1B or more (iab.com, January 2026).
For volume, the strongest published dataset is Motion’s Creative Benchmarks 2026, covering 578,750 unique creatives across 6,015 advertiser accounts and $1.29 billion in realized Meta spend, for creatives launched between September 2025 and January 2026. Around half of creatives are switched off before 28 days. Only 4% to 8% qualify as winners, depending on account size, and mid-tier accounts ship roughly six to seven creatives a week.
Motion defines a winner as spend at least ten times the account median with a $500 floor, and measures success by realized spend rather than by return on ad spend (Motion methodology, October 2026). It is a vendor dataset from its own connected accounts, so it may skew toward sophisticated advertisers. The methodology is published and specific, which puts it well ahead of the unsourced figures above.
The implication for pack sizing is important. If roughly one creative in 12 to 25 becomes a winner, and mid-tier accounts sustain six to seven new creatives a week, the winning pattern is a steady stream rather than one large batch. A pack fills a week. It does not fill a quarter, and sizing it as though it should is how the credit budget disappears in the first two weeks.
How to build the pack
- Set the cell count from the media plan. Weekly conversions divided by roughly 50 gives the ad sets that can stabilize. That is how many genuinely different creatives you need.
- Pick the model against that number. If the plan supports twelve cells on a Premium budget, pick a model where twelve clips, times your takes, fit the month. Multiply before you run.
- Start from the product photo. Assets made from your real product keep the fixed elements in place. Six rules for AI product video cover what makes a product clip work.
- Fix the source ratio before generating. If the pack is vertical and you are using image-to-video, the input image needs to be 9:16 already, because some models inherit it.
- Decide sound at the model layer. Native audio can change the cost of a clip, and on Kling 2.6 it also needs 1080p.
- Launch the cells together and leave them alone. Adding ads to a live ad set restarts its learning phase, which is the argument for batching in the first place.
- Save the pack as a workflow. A saved workflow runs the same structure on the next product or drop, so nobody rebuilds it.
- Disclose the AI where the platform requires it. Building TikTok video ads with AI covers the current specifications and disclosure rules.
For the wider spend picture around all of this, scaling AI ad campaigns without burning budget covers what Meta does and does not publish about creative volume. A steady stream of packs also raises the question of who builds them, which how small agencies scale creative production without hiring answers for a small team.
DesignerBox is AI creative production for brands and agencies. The pack size decision and the model decision happen in the same place. You build the pack once as a workflow, with your brand, your product and your rules. You set the brand once, and the workflow reads it on every run. Batch runs one workflow over a whole sheet of products. In DesignerBox, the AI video ads page covers the clip from a product photo to a finished ad.
The Ad versions with new headlines template makes new variants from an ad that already worked. The Ads from a product photo template covers the still variants in the same pack. The clips still need cutting and captions. The video editor is a timeline with several tracks. It adds captions from speech, a voiceover, music and transitions. Stills and clips come from one account and one set of brand rules.
To build the first pack, start from a template, add your product and your brand, and run it.
FAQ
What is an AI video ad pack?
An AI video ad pack is one batch of video ad variants generated from a single brief and launched together as a structured test. Variants normally differ by hook, format and audience. The pack is a production unit rather than a delivery unit, so it has to be split across ad sets, each with its own budget, before it runs.
How much does an 18-clip AI video ad pack cost?
The model decides most of it. Multiply the cost of one clip by 18, then by your takes per clip. One 8-second clip in DesignerBox costs 40 to 560 credits, depending on the model, and the total is shown before the run. That spread is why model choice sets the maximum pack size.
How many video ad variants should I put in one pack?
As many as your weekly conversion volume can drive to Meta’s learning threshold of about 50 results per ad set, and no more than your credits allow at your chosen model. Meta’s own ceiling is no more than five ads per ad set for optimal delivery (facebook.com, September 2026).
What should I vary between video ad pack variants?
Vary the opening three seconds, the narrative structure, the setting, the pacing and the call to action. Hold the product, brand palette, type, logo treatment and offer fixed. Meta’s creative diversification guidance asks for assets that are truly different in look, feel, storyline and message. Variants that differ only in their first two seconds do not meet that.
Which model is cheapest for a video ad pack?
Run one test clip on two or three models before you commit the pack. Cheaper models trade resolution and motion quality. A large pack means the lighter models, and the premium look means a smaller pack.
How often should I refresh a video ad pack?
Refresh when results decline. TikTok’s video advertising guide says to refresh creatives every 2 to 4 weeks, or sooner if click-through or view-through rate declines (ads.tiktok.com, October 2026). Its creative best practices page ties refresh to signals: delivery results that decline consistently, or falling daily new users (ads.tiktok.com, October 2026). Meta publishes no fixed refresh interval. Neither company gives seven days, so the widely quoted seven-day rule has no documented origin in their guidance. Motion’s 2026 dataset suggests mid-tier accounts sustain around six to seven new creatives a week.
Can one product photo produce a whole video ad pack?
Yes, and it is the approach that keeps the fixed elements fixed. Every variant comes from the same source image and brand profile, so the product, palette and logo stay consistent across the pack. Someone still reviews every clip before it runs. Image-to-video inherits the source ratio on some models, so set the source to 9:16 first if the pack is vertical.
Sources
- Meta’s learning threshold of about 50 results per ad set in the week after the last significant edit: Meta Business Help Center, about the learning phase, accessed 2 October 2026
- Meta’s creative diversification guidance on assets truly different in look, feel, storyline and message: Meta for Business, December 2025
- Meta’s recommendation of no more than five ads per ad set for optimal delivery: Meta for Business, ad set structure, accessed September 2026
- The “six or fewer creatives per ad set” line, verified from an archived capture and reported here as removed rather than current: Meta ad volume page, February 2025
- TikTok’s 3 to 5 unique creatives per ad group, 3 to 5 ad groups per campaign, and its signal-based refresh guidance: TikTok Ads Manager Help, creative best practices, page last updated June 2025, accessed 2 October 2026
- TikTok’s advice to refresh creatives every 2 to 4 weeks, or sooner if click-through or view-through rate declines: TikTok for Business, Video Advertising Guide for Small Businesses, read 7 October 2026
- Creative win rates, 28-day switch-off rate and weekly creative volume across 578,750 creatives and $1.29 billion in Meta spend: Motion, Creative Benchmarks 2026, with the pages portfolio breakdown by tier, scale and volume and methodology, accessed October 2026
- The 86% “using or planning to use” figure: IAB Digital Video Ad Spend and Strategy report (iab.com, July 2025)
- 83% of ad executives reporting AI deployed in the creative process, from 104 US ad executives: IAB, The AI Gap Widens, January 2026
- Veo 3.1 clip lengths, always-on audio and the 8-second requirement for 1080p and 4K output: Google Gemini API Veo documentation, accessed September 2026
- Kling 2.6 clip lengths, the 1080p requirement for audio, and the missing aspect ratio parameter on image-to-video: Kling capability map, accessed September 2026
- Seedance 2.0 clip length, optional audio flag and 4K output: BytePlus ModelArk documentation, accessed October 2026
- Runway Gen-4.5 clip length range and no documented audio output: Runway API documentation, accessed October 2026
- Runway’s December 2025 announcement of native audio for Gen-4.5: Runway Research, accessed October 2026
- DesignerBox plan gates and the 8-second video cost range: DesignerBox pricing page (designerbox.ai/pricing), September 2026
Meta learning phase and creative diversification guidance verified against Meta Business Help Center and Meta for Business pages, July 2026. The learning phase page was re-checked on 2 October 2026. The “six or fewer creatives per ad set” line is verified from an archived February 2025 capture of Meta’s ad volume page and is reported here as removed rather than current. TikTok creative volume and refresh guidance verified against TikTok Ads Manager Help, on a page last updated in June 2025 and re-checked on 2 October 2026. Model capabilities verified against Google’s Gemini API Veo documentation, BytePlus’s Seedance documentation, Kling’s capability map and Runway’s API documentation, all accessed September 2026. The Seedance and Runway pages were confirmed again in the 2 October 2026 fact check. Creative volume benchmarks from Motion’s Creative Benchmarks 2026, a vendor dataset of 578,750 creatives and $1.29 billion in Meta spend with published methodology. TikTok’s video advertising guide was read on 7 October 2026. Frequency thresholds, the seven-day refresh rule and per-variant conversion minimums are deliberately excluded because Meta’s and TikTok’s documentation does not support them. DesignerBox plan allocations are current as of September 2026. Individual results vary.
