Breaking

PYLER Uses NVIDIA to Scan 2M Daily Videos for Ad Brand Safety

June 25, 2026 at 16:00 EDT

Sales & Business
Media Generation
Infra & Chips

South Korean video-understanding firm PYLER is analyzing two million videos a day on NVIDIA-accelerated infrastructure to improve brand safety and campaign performance for advertisers, in a case study highlighted by NVIDIA.

PYLER · NVIDIA AI Case Study

AI that watches 2 million videos a day — and cuts brands' exposure to unsafe content by 76.7%

Built on NVIDIA's full stack, PYLER's "Antares" model reads visuals, audio, text and metadata in near real time — collapsing a 30-second human review into 11 milliseconds.

76.7%

Reduction in ad exposure to unsafe content

Videos analyzed daily (60 years of footage in 24h)

~10×

Lower operating cost vs. prior approach

99.9%

Efficiency gain: 30s review → 0.011s

Verification time, drawn to scale

Same task, same unit (seconds) — human review vs. Antares.

30 s

Human review

0.011 s

Antares (AI)

≈ 2,700× faster — the green sliver is barely a hairline next to the orange column.

Blackwell upgrade: throughput gains

Multiples vs. the previous generation (1× baseline).

1×

Prior gen

3×

Multimodal training

4×

Video preprocessing

5×

Hyperparameter search

Model training cycles shortened from 3 months → 1 month.

Proof in live campaigns

Nongshim · "KPOP Demon Hunters"

Unsafe budget share fell in one month

19.3% → 4.67%

Kenvue · Tylenol

Reviewer cost reduction

99.963%

How Antares understands a video

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.