Local SDK speed
46.2x
Mean local SDK throughput advantage over generic vector search.
Product spotlight
XLerate™
DNA
PhotoDNA3 provides a high-trust signal robust to real-world image changes. XLerate DNA makes that signal searchable at scale.
AWS efficiency
8.1x
Mean throughput multiple over generic vector search.
Slash cloud cost
6.8x
Mean AWS infrastructure cost reduction over generic vector search reaching for the same performance.
Uncompromising
100%
Engineered to always match and never false positive within your parameters - proven on millions of vectors.
PhotoDNA meets XLerate
Purpose-built for high-trust image matching.
PhotoDNA is designed to find resilient matches across real-world image changes. XLerate DNA is designed to serve those matches at scale.
Generic vector systems are built for similarity search, where “close enough” is the contract. With these systems, you can gain more accuracy by sacrificing speed - which also increases operational time and thus costs - or speed for accuracy. But PhotoDNA workflows are different. Trust & safety and forensic teams need certainty, speed, and cost control at the same time.
XLerate DNA removes that tradeoff: full-recall PhotoDNA search, GPU-class throughput on commodity CPUs, and an performance that grows stronger as datasets grow.
Exact matching leaves PhotoDNA's power on the table.
Brute force does not scale.
Generic vector search was not built for this.
XLerate DNA was.
Built around certainty
PhotoDNA workflows cannot afford “close enough.” XLerate DNA is engineered to preserve the recall contract while serving matches at operational speed.
GPU-class speed on CPUs
XLerate DNA brings high-throughput PhotoDNA search to ordinary CPU infrastructure, reducing dependence on expensive GPU serving stacks.
Bigger databases should not mean weaker search
As PhotoDNA datasets grow, XLerate DNA is designed to not only maintain matching confidence, but database growth improves both recall and speed.
Deploy without retraining the team
XLerate DNA works with PhotoDNA hashes directly, giving teams a faster serving layer without asking them to redesign their workflows.
Hashes and vector space
Image changes get lost in vector space.
Move the sliders and watch the edited image drift away from the original. That drift is why PhotoDNA needs search built for high-dimensional vectors.
Reference image
The original image and the baseline 144-byte hash already stored in the dataset.
Baseline 144-byte hash
Candidate image
The same scene after small changes that move the hash through vector space.
Candidate 144-byte hash
L2 0.000
XLerate advantage
XLerate DNA makes PhotoDNA vector search operational.
A simple transformation can move the hash. That is not a failure of PhotoDNA - that's its strength. The hash isn't adrift in space: PhotoDNA is precisely creating the signal needed to keep track of the image under change. It's up to us to follow it accurately.
The hard part is serving that at scale. Exact matching is cheap, but brittle, and leaves all the power of PhotoDNA on the table. Brute force keeps the whole search space alive - which becomes too expensive to operate. Generic vector search works, but treats recall as an ideal rather than mission-critical - all of which scales poorly in both cost and performance.
XLerate DNA changes the operating model. Its proprietary, asymmetrical approach2 for PhotoDNA vector space preserves the recall contract while delivering the throughput and economics teams need in production.
XLerate™ DNA
The hash changes
XLerate DNA tracks the signal, and successfully matches the new hash to the original.
Intelligent growth
Only an optimal amount of hashes for an image are retained - capped growth that makes searches shallower.
Speed increases
As coverage of matched images accumulate, searches become faster without surrendering recall.
Companion page
Measured performance
See the benchmark story in full.
The benchmark page takes a deep dive into the performance characteristics of XLerate DNA including local SDK throughput, AWS single-query throughput, p95 latency, cost parity, stress behavior, and the deployment story behind the service results.
Open the benchmark summary