Talking-head format
19%
58 of 306 were aged 60+ days, compared with 31 of 808 graphics, or 3.8%.
Try this: open with one specific business problem, teach a useful step, then give a relevant next action.
Ad Lab / Jev for marketing
Give it a batch of ads. Find the hooks, formats and offers they use, then click through the examples behind each pattern. Get the full app and setup so you can research a competitor, a category or your own creative library.
By Artem Novitckii · October 27, 2026 · Free code + full setup
What's inside
The saved Hormozi snapshot
The saved sample contains 1,214 ad records collected on October 12, 2026. These comparisons describe the share of dated active ads that started at least 60 days before collection. They give you examples to study and hypotheses to test.
19%
58 of 306 were aged 60+ days, compared with 31 of 808 graphics, or 3.8%.
Try this: open with one specific business problem, teach a useful step, then give a relevant next action.
18.6%
22 of 118 were aged 60+ days. 16 of those 22 used numerical proof.
Try this: lead with an outcome you can substantiate and show the evidence behind it.
17.3%
56 of 324 were aged 60+ days. 38 of those 56 used a talking-head format.
Try this: name who the workshop is for, what they will learn and the registration step.
The sample-wide share was 7.9% (96 of 1,214). The archive's historical labels were generated with the earlier provider and have not been relabelled as new Jev results. The current app uses Jev's native image API. Active age does not establish continuous delivery, profitability or future results.
01 / What you are building
Bring your images, videos or JSON. Optionally collect public Meta Ad Library records with Apify.
It receives the images directly, plus copy and any transcript, and returns seven typed classifications.
Filter patterns, inspect original ads and turn a useful structure into a new brief for your brand.
The labels cover format, hook, angle, offer, proof, objection and CTA. Jev does not receive campaign age or outcome metrics for classification. The dashboard joins the labels to the available collection metadata afterward.
02 / Install once
package.json.git clone https://github.com/artemnovitckii/ad-pattern-lab.git
cd ad-pattern-labnpm ci
npm run setup
npm startThe no-key demo includes fictional creatives, findings and briefs. Replay animates those stored fixtures without paid requests. Stop the server with Ctrl+C when you are ready to edit .env.
Install FFmpeg for image preparation and video sampling. Both ffmpeg and ffprobe must be available in your terminal.
brew install ffmpegwinget install --id Gyan.FFmpeg --exactClose and reopen PowerShell afterward.
sudo apt update
sudo apt install ffmpegOther builds are linked on FFmpeg's download page. No separate OCR package or language-model download is required.
03 / Connect only what you need
Open the .env file created by setup. Fill in your own values:
ANALYSIS_PROVIDER=jev
TYPESAFE_API_KEY=your_typesafe_key
JEV_MODEL=jev-latest
APIFY_TOKEN=your_apify_token
TRANSCRIPTION_PROVIDER=none
PORT=5194Leave APIFY_TOKEN empty if you are importing your own files.
TRANSCRIPTION_PROVIDER=fireworks
FIREWORKS_API_KEY=your_fireworks_keyTRANSCRIPTION_PROVIDER=groq
GROQ_API_KEY=your_groq_keynpm run doctor
npm startDoctor checks tool availability and key presence without making paid calls. It does not verify model access. Keep keys in your local .env, which Git ignores, and restart after changing it.
04 / Your first real batch
Requested collection counts are not a promise of complete account or market coverage. Public availability and expired media URLs can limit the batch. Completed matching evidence is cached; saved replay does not rerun the providers.
05 / Turn the findings into your next test
Filter by format and hook. For example, compare talking-head ads with result-led openings against the other creatives in the same batch.
Use Explore examples to browse the matching ads. Click each thumbnail to see its opening, body, CTA, seven labels, available image evidence and original ad link. Copy sections are ad copy, not an invented spoken transcript.
The creative landscape groups similar labels. The alternate map compares age and observed ad IDs per family. These are research views, not ROAS, profitability or predicted survival odds.
Supply your brand, audience and verified claims to create editable brief outlines. You can add an optional chat-model writer for prose drafts. Jev reviews clarity, brand fit and claim support; your real campaign test determines what works.
Use Replay in the main dashboard for your saved results, or open /variations.html for no-key simulation layouts. Wall, counters and map share a clock. The timer is playback time, not measured API speed. Pause with Space, restart with R and toggle controls with C.
06 / If something gets stuck
Check that .env is beside package.json, the key is named TYPESAFE_API_KEY, and the file is not .env.txt. Confirm access and billing in TypeSafe, save, then restart the app.
Open the error and saved evidence. Public media URLs can expire; upload the creative or collect fresh media. Four sampled video frames can miss the message. Add an actual transcript or factual notes when needed.
Check quota or temporary capacity, wait, then resume unfinished work. Lower ANALYSIS_CONCURRENCY if needed. A lost paid Jev response is not automatically retried because it may already have been billed.
Check FFmpeg and retry extraction. The app converts images to JPEG and prepares them for TypeSafe's size limits. Native image support is experimental, so check TypeSafe's current documentation if the error persists.
Add your brand, audience and verified facts, then request brief outlines. Connect a chat-model writer only if you want prose drafts. Jev classifies and reviews; it does not write those drafts itself.
Find the existing run ID in Apify Console and attach that run. Do not start another paid collection for the same job. A dollar cap may stop the Actor before the requested count.
Jev receives creative images, copy and any transcripts. Apify receives collection inputs; an enabled speech provider receives audio. Keys remain server-side. Local uploads, frames and caches live in the ignored data folder. Review exports before sharing.
Get Ad Lab, add your keys, and study the examples before deciding what to make.
Enter your email to unlock the app, setup guide, saved research and the rest of my resource library. You'll also get my weekly email. Unsubscribe anytime.
Moshn / AI automation agency
Moshn builds AI systems around the tools your team already uses, from lead qualification and follow-ups to onboarding and reporting.
Book a call. Tell us what your team is doing manually, and we'll show you what can be automated.