NEURAL CREATIVE ID

by ShookAI

Recognizing ad creatives by what they look like — not how they're encoded

THE PROBLEM

Ad creatives are permuted endlessly — same creator, same template, different hook or CTA. To analytics systems, these look like completely unrelated ads.

🎬
Hook A + Body + CTA
SHA-256
a7f3c…
🎬
Hook B + Body + CTA
SHA-256
2e91b…
🎬
Hook C + Body + CTA
SHA-256
f04d8…

Same template, same creator — every hash is completely different

THE INSIGHT

Don't ask "are these the same file?"

Ask "do these look like the same ad?"

TRADITIONAL
SHA/MD5 — any change breaks it
TLSH/ssdeep — byte-level, codec-dependent
pHash — pixels only, no semantics
NEURAL
Understands content, not encoding
"Person talking to camera" ≈ "person talking to camera"
Survives any re-encoding or overlay

HOW IT WORKS

01
🎞️
SAMPLE
Extract random frames
from the video
→
02
🧠
EMBED
AI model converts each
frame to a numeric vector
→
03
🔍
MATCH
Compare vectors to find
similar creatives instantly

No modification to the creative. No segmentation needed. Works on any video already analyzed.

WHAT IT ENABLES

CREATIVE FAMILIES
Template A
Hook v1
Hook v2
CTA var
Recut
ANALYTICS UNLOCKED
Aggregate performance across all variants
Compare: which hook variant wins?
Attribute templates to creators
Detect when a winner spawns copies
Funnel attribution across the family
🧠

SAME AD. DIFFERENT FILE.

NEURAL ID SEES THROUGH IT.

No modification to creatives. Works retroactively.
Identifies template families across any platform.