An AI video ad pack is a single batch of video ad variants built from one brief, split across hooks, formats and audiences. Guides size these packs with a fixed grid, usually three hooks by three formats by two audiences. For video that grid breaks on cost before it breaks on anything else. Eighteen clips at a premium model costs fourteen times an Ultra plan’s entire monthly allocation, so model choice and pack size are one decision.
You have seen the grid. Three hook families, three formats, two localisations, eighteen assets, ship it Monday. It is a tidy number, and for static creative it is at least arguable. For video it collapses the moment anyone prices it.
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 inverts that. Video is priced per second of output, it is the single most expensive operation in AI creative production, and an eighteen-clip pack can cost more than a year of your plan before a single impression is served. The binding constraint arrives earlier, and it arrives in your credit balance.
This covers what a video ad pack costs across six models, 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
- For video, cost binds before measurement does. An 18-clip pack at Veo 3 with audio costs 115,200 credits, 14 times the Ultra plan’s monthly 8,000. The same pack at Seedance Pro Fast costs 2,700.
- Model choice and pack size are one decision, not two. Picking the model first and the count second is how teams end up with a batch they cannot afford to finish.
- Meta publishes no recommended number of ads per ad set. The “six or fewer” line that circulated in this category 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 (facebook.com, July 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, June 2025).
- Kling 2.6 is silent at 720p. Sound requires 1080p professional mode, so a sound-on pack costs more per second than the headline rate suggests.
- 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, Creative Benchmarks 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.
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.
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 actually 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 stabilise, and that is your cell count.
The second question is where video separates from static, and it is the one nobody in this category prices.
What an AI video ad pack costs in credits
Video is priced per second of output. Pack size therefore multiplies against the most expensive operation in AI creative production. In DesignerBox the numbers look like this.
| Model and settings | 6-clip pack | 12-clip pack | 18-clip pack | Share of Ultra’s monthly 8,000 |
|---|---|---|---|---|
| Seedance Pro Fast, 720p, 5s | 900 | 1,800 | 2,700 | 34% |
| Kling Standard, 720p, 5s | 1,350 | 2,700 | 4,050 | 51% |
| Sora 2, 720p, 8s | 9,600 | 19,200 | 28,800 | 360% |
| Veo 3 with audio, 8s | 38,400 | 76,800 | 115,200 | 1,440% |
An 18-clip pack at Veo 3 with audio costs more than fourteen times the entire monthly credit allocation of the $200 Ultra plan. The same eighteen clips at Seedance Pro Fast fit inside Ultra with room to spare, and land just above Premium’s 2,500.
That spread is four hundred times wide. No other decision in the pack moves the number that much, which is why model choice is not a quality preference to settle later. It sets the maximum pack size you can build.
Plans run from free with 112 credits, through Basic at $15 for 500, Pro at $35 for 1,000, Premium at $75 for 2,500, and Ultra at $200 for 8,000. AI video starts at Premium and the commercial licence starts at Pro. A still image is 5 credits, which is why packs that carry still variants alongside clips stretch a budget considerably further. The full per-model breakdown is in what each AI video model costs per second.
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 generating anything.
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 bite. Every spec below was verified against the provider’s own documentation in July 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 |
| Sora 2 Pro | Up to 20 seconds per generation | Yes, synced | Yes, 1080x1920 |
| 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 outputs silent video at 720p. Sound requires the 1080p professional mode, so a sound-on Kling pack is a more expensive generation than the standard rate implies (app.klingai.com, May 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, not after.
Veo 3.1 requires 8-second clips for 1080p or 4K output (ai.google.dev, January 2026). A pack of 6-second Veo clips is a 720p pack, and 8 seconds is where the credit cost climbs fastest.
Runway’s published API schema for Gen-4.5 carries no audio parameter, while its schemas for Veo and Seedance both do. Treat Gen-4.5 as silent unless you confirm otherwise in the product.
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, which it distinguishes from creative iteration and manual refinement (facebook.com, 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 recognisably 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 derives from the same product photo and the same brand profile, consistency stops depending on anyone remembering to check it, and nothing comes out looking generic AI. That mechanic is covered in keeping brand consistency across AI creative.
Where the numbers in this category come from
Search for pack sizing and you will find confident figures with no origin. Worth knowing which is which before planning against them.
| 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. TikTok ties refresh to declining delivery, with no day count |
| ”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) |
That last one circulates constantly as a current statistic. It 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, from 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 realised 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 realised spend rather than by return on ad spend. It is a vendor dataset from its own connected accounts, so it skews toward sophisticated advertisers, but the methodology is published and specific, which puts it well ahead of the unsourced figures above.
The implication for pack sizing is worth sitting with. If roughly one creative in ten to thirteen 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 fortnight.
How to build the pack
- Set the cell count from the media plan, not the grid. Weekly conversions divided by roughly 50 gives the ad sets that can stabilise. 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, the pack is Seedance Pro Fast at 5 seconds, not Veo with audio. Multiply before you generate.
- Start from the product photo, not a prompt. Assets derived from your real product hold the fixed elements automatically.
- 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 is where per-second cost rises hardest, and on Kling it also forces 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. The next product or drop reruns the structure instead of rebuilding it, which is where batch production actually pays back.
- 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.
DesignerBox runs this end to end from one product photo. Every top image and video model sits behind one subscription, so the pack size decision and the model decision happen in the same place, and packs save as workflows you rerun for the next product. The Social Media Ad Studio handles the batch, and the Marketing Studio covers the wider campaign.
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?
It depends almost entirely on the model. In DesignerBox, 18 clips of Seedance Pro Fast at 720p and 5 seconds costs 2,700 credits. The same 18 clips at Veo 3 with audio and 8 seconds costs 115,200 credits, more than fourteen times the Ultra plan’s monthly allocation of 8,000. That four hundred times 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 publishes no recommended number of ads per ad set, so any fixed figure quoted elsewhere is a practitioner convention rather than platform guidance.
Does Meta recommend six or fewer ads per ad set?
Not on the live help centre. That sentence appeared on Meta’s ad volume page as of February 2025 and is no longer published there, which is why frameworks built on it are quoting an archived figure. Meta’s current published position names no number of ads per ad set.
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, not variants that differ only in their first two seconds.
Which model is cheapest for a video ad pack?
Of the options priced above, Seedance Pro Fast at 720p and 5 seconds is the lowest at 150 credits per clip, against 6,400 for 8 seconds of Veo 3 with audio. Cheaper models are shorter and quieter, so the honest trade is that a large pack means shorter, silent clips, while sound and length mean a small pack.
How often should I refresh a video ad pack?
TikTok’s published guidance is signal-based rather than calendar-based, triggering refresh when delivery results decline consistently or daily new users fall (ads.tiktok.com, June 2025). No platform publishes a fixed refresh interval, so the widely quoted seven-day rule has no documented origin. 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 derives from the same source image and brand profile, so the product, palette and logo stay consistent across the batch without manual checking. Image-to-video inherits the source ratio on some models, so set the source to 9:16 first if the pack is vertical.
Meta learning phase and creative diversification guidance verified against Meta Business Help Center and Meta for Business pages, July 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, June 2025. Model capabilities verified against Google’s Gemini API Veo documentation (January 2026), OpenAI’s video generation guide, Volcano Engine’s Seedance documentation (July 2026), Kling’s capability map (May 2026) and Runway’s published API schema (February 2026), all accessed July 2026. 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. Frequency thresholds, fixed refresh intervals and per-variant conversion minimums are deliberately excluded because no platform documentation supports them. DesignerBox credit costs and plan allocations are current as of July 2026. Individual results vary.
