Pencil vs Omneky for AI ad creative testing
For AI ad creative testing, Pencil suits teams that want many variants from several AI models, a predicted score before launch and tracking across seven channels, from $14 a month. Omneky suits teams that want image, short video and avatar variants tested on up to nine channels, with tagged creative insights and a built-in budget step, from $29 a month. Neither tool decides what a fair test is for you, so set the budget per variant and the rule for calling a winner before you generate anything.
How do Pencil and Omneky support creative testing?
Creative testing has three parts: making enough distinct variants, running them under fair conditions, and reading the results in a way that tells you what to make next. Both tools cover the first part well and parts of the second and third.
Pencil brings image, video and copy models into one workspace, gives "Realtime AI insights and Scores" on every plan, and lists launch and tracking on Facebook, Instagram, TikTok, YouTube, Google Display, DV360 and LinkedIn (pricing page). Its strength is variant breadth across models under one set of brand rules.
Omneky runs a plan, create, allocate and launch flow. It makes image ads in 12 ratios, short cinematic video, and on Standard and above, commercials and UGC avatar video. Every plan includes its Agent, Automations and Insights with a drilldown view, and its home page says it launches on nine channels, including Meta, Google, TikTok and OpenAI Ads (omneky.com). Its strength is format breadth plus reporting in the same tool.
Prices and features below come from each vendor's own site, checked 24 October 2026.
We make Nagent. Our creative strategy lead, HOOK (Beta), works on the angle and the read rather than bulk generation, and has its own section near the end.
What does each one cost for a testing programme?
Pencil
What it is: a multi-model generative AI platform for marketing teams with brand safety controls, scores and ad launching.
Best for: testing many copy and visual variants of one concept, especially across brands or markets, and teams that want enterprise terms later.
Pricing (checked 24 October 2026, trypencil.com/pricing): Core $14 a month billed monthly or $11 billed annually, with 50 generations a month and one workspace. Growth $55 or $44, with 250 generations and unlimited workspaces. Pro is custom, with unlimited generations, feed-based bulk generation and unlimited users, brands and markets.
Testing limit: 50 generations a month on Core covers a small weekly test at best once you count retries. A steady programme usually means Growth.
Omneky
What it is: an AI creative platform for image, video and avatar ads with launching and tagged creative insights.
Best for: testing format against format (static vs short video vs avatar), and teams that want to launch and read in one place.
Pricing (checked 24 October 2026, omneky.com/pricing-plans): Lite $29 a month billed monthly or $24 yearly, with 200 credits, one brand and one seat. Standard $99 or $79, with 1,000 credits, three brands and three seats. Pro $249 or $199, with 4,000 credits. Enterprise is priced with sales. Top-ups cost 10 credits per dollar and never expire.
What testing volume costs: a 1K image ad is 5 credits and a short commercial up to 15 seconds is 30, so Lite's 200 credits make about 40 image ads, and Standard's 1,000 make about 200 image ads or 33 short commercials, by the page's own figures. Connecting an ad account costs no credits.
Testing limit: video and avatar tests start on Standard, because commercials, UGC avatar and clone video are not on Lite.
Which one should you pick for testing?
Pick Pencil if your tests are mostly copy and visual variants of a known format, you want several models to compare, or you manage several markets and want one workspace each.
Pick Omneky if you need to test static against video against avatar, want per-format credit costs you can budget, and want insights tagged from the creative itself.
Pick neither yet if you cannot say what each variant is testing. Generating 50 ads without a hypothesis gives you 50 data points that teach you nothing.
How do you design a fair creative test?
The tool makes the variants. The test design is yours, and it decides whether the result is worth anything.
Write one hypothesis per test. "A texture-led hook beats a discount-led hook for cold audiences" is testable. "Try new ads" is not.
Change one thing at a time. Hook, format, offer or presenter. If two change, you will not know which one won.
Decide the winning metric before launch. Cost per purchase or cost per lead, not CTR, unless clicks are the goal.
Set a minimum sample. Pick a fixed number of conversions per variant before you will call it, and work backwards to the budget. For example, at a $40 cost per purchase and a threshold of 30 purchases, each variant needs about $1,200 of spend.
Set a kill condition for each variant before it runs, such as "pause if spend passes twice the target cost per purchase with no purchase".
Keep audiences and placements equal, or use the platform's own experiment tool so the split is clean.
Write down what won and why in a place the next brief can read.
That last step is where most AI testing programmes fail. The variants change every week, but the lesson stays in one person's head. Our guide to turning ad performance data into the next creative brief walks through it.
What do Pencil and Omneky not tell you about testing?
We read what both publish on testing. Omneky's comparison of campaign automation platforms says its workflow feeds performance data back into the next round of generation. Pencil's pages point to scores and predictions. Neither page we read explains:
How a score relates to your own results. A pre-launch score is a prediction. Ask how often it has matched in-market winners on accounts like yours.
How much budget each variant needs before a result means anything.
Who is allowed to move money. Omneky's home page demo shows an agent moving budget to the channels that returned most before a "Confirm launch" step; the demo is labelled illustrative. Find out exactly which actions each tool can take in your ad account and who approves them.
What happens to claims. Testing many variants multiplies the number of claims you make. One unsupported claim in a winning ad can cost more than the test saved.
What questions should you ask on a Pencil or Omneky demo?
Does a retry or a resize use a generation or credits?
Which channels can I launch to on the plan I would buy, today?
When the tool launches, can it change budgets or bids, or only create ads?
How are winners called in your insights: by which metric, over what window, against what baseline?
Can I export the creative tags and results to my own sheet or BI tool?
What brand and claim controls exist before an ad goes live?
Where does Nagent fit?
Nagent is multiplayer AI for growth: people and AI agents in one shared context, delivering the number. HOOK, in Beta on the Marketing team, covers the strategy side of testing that generators leave to you:
Bets with kill conditions. ANGLE writes a creative brief as a set of bets. Every bet carries a structured kill condition, and each opens a concept you can follow from bet to winner.
Hooks and scripts per channel pack. SCRIPT writes hook sets and scripts for Meta cold 9:16, Meta retargeting 1:1, TikTok Spark and Google RSA.
Claims checked. GATE checks every claim against the brand's proof and holds what it does not support.
Winners called against your own ads. PULSE reads a connected Meta account or an uploaded .csv or .xlsx export from any platform, ranks each ad against your own ads in its channel pack over 90 days, and says when a kill condition is met. VAULT turns the results into a playbook the next brief reads.
What it does not do: HOOK makes read calls only. It cannot launch, pause or change a budget, and a person pauses the ad when a kill condition is met. It does not mass-produce statics; RUPA renders those from HOOK's brief. If you want budget changes proposed on Google or Meta, NIA proposes them and every one waits for a person to approve. Nagent work is metered in credits; see pricing.
Frequently Asked Questions
Is Pencil or Omneky better for Meta ad creative testing?
Both list Meta among their launch channels. Omneky suits format tests (static, short video, avatar) with per-format credit costs, starting at $29 a month. Pencil suits variant tests across several AI models at a lower entry price of $14 a month. Prices checked 24 October 2026.
How many ad variants should you test at once?
As many as your budget can take to your minimum sample per variant. If each variant needs about $1,200 of spend to reach your threshold, a $6,000 weekly test budget supports about five variants, not fifty.
Does AI creative testing automatically pause losers and move spend?
Some tools can act in your ad account; Omneky's demo shows a budget step before launch, and Pencil lists launch and tracking. Many teams keep a person on pauses and budget moves. On Nagent, HOOK flags met kill conditions and a person pauses the ad.
What does Pencil vs Omneky cost for a brand spending under $50k a month on ads?
Tool cost is small next to media at that level: Pencil Growth is $55 a month and Omneky Standard $99 a month billed monthly (checked 24 October 2026). The bigger cost is test media, so size the testing budget first.
See how HOOK turns test results into the next brief. Book a demo.