Why we built Advar.

I'm Yibin, founder of Advar. Over the past eight years, I've led video recommendation systems for apps with tens of millions of monthly active users, worked on e-commerce SaaS platforms connected to billions of dollars in GMV, and served as an AI product manager at a public company building AI training platforms.

In 2026, as AI-generated video ads started moving from novelty experiments to real production, I saw a rare overlap between the systems I had built and where performance marketing was heading. Creative testing is the single highest-leverage variable in paid social. Yet when growth teams first experimented with AI, the results were underwhelming—leading many to wonder: does the algorithm secretly penalize AI-generated ads?

Having worked on the ranking side of recommendation systems, I knew that wasn't how it works. Social recommendation engines and paid auctions (from Meta Advantage+ to TikTok's For You feed) don't punish an ad for being generated by AI. They punish it for triggering instant ad-blindness.

When a video relies on recycled templates or public digital avatars that hundreds of brands are already running, viewers swipe away in the first second. Thumb-stop rates collapse, watch time evaporates, the auction quality score plummets, and the CPM required to buy reach skyrockets. The idea didn't fail; it simply blended into the feed's digital wallpaper.

To win in the feed, quality and volume both matter:

• Quality earns the thumb-stop, wins cheaper CPMs in the auction, and lets you scale ad spend without burning out margins.
• Volume gives advertisers enough shots on goal to test hooks, angles, and offers inside a black-box market.

Yet the tools on the market forced teams to choose one:

• Node-based workflows and raw models: Skilled AIGC creators and specialized studios can craft visually impressive pieces, but the process is heavily manual. Even a dedicated expert can only produce a handful of launch-ready ads a day—far too slow to sustain systematic testing at scale.
• Templates and lip-sync avatars: They can generate hundreds of variations a day, but the output feels stiff and repetitive, with no genuine product demonstration and zero creative range. Audience fatigue begins on day one.

Advar was built to solve that tension: bringing studio-grade visual craft and high-velocity iteration into a single automated pipeline.

Too many AI ad tools claim they will replace the human marketer. I think that misses where the real difficulty sits.

Rendering pixels was never the hardest part of a winning ad. The hard part is understanding customer friction: the unvoiced objection, the emotional trigger, the reason someone scrolled past the last ten offers. No model has sat in your customer interviews. That strategic judgment stays yours.

Whether you are a DTC brand, a performance creative agency, or an e-commerce operator, the fundamental challenge is always the same: how to turn the ideas in your head into launch-ready ads—at scale, and at high quality. Advar exists to be that production leverage, so your best creative concepts actually make it to the auction.

If you have feedback, or want Advar to better fit your team's workflow, I'd be glad to hear from you directly. I read every message.