And we’ve published peer-reviewed research at top AI conferences E.g. CVPR, NeurIPS, ECCV, AAAI, Interspeech which are available at https://www.realitydefender.com/research
This is something we're constantly updating, upgrading, iterating, and improving on. Every. Single. Day.
Whether it's introducing new models, deprecating old ones, or improving existing ones, there is an element of both staying current but also looking ahead at research. Many of the new models generating hyperreal content we catch on day one because they're based on existing technology and/or research.
We'd rather not tip our hand on any/all techniques used to discern actual users from bad actors and those seeking to reverse engineer, but suffice to say we do have are methods (and plenty of them).
As noted elsewhere, we give confidence scores between 1-99%. We also use many different models for each modality for a more robust and complete answer with each scan, and each model has its own confidence score.
I understand this is in jest, but unfortunately AI generation tools more or less stopped the six-finger issue a couple of years ago. We are decidedly not a model used for the express detection of finger abnormalities, but a multi-model and multimodal detection platform — driven by our Public API (which you can try for free right now, btw) — which uses many different techniques to differentiate between content that is likely manipulated and likely not manipulated.
Thank you. As an inference-based detection platform, our models go into every scan with the assumption that all files are both not the original/ground truth AND the files have been likely transcoded. We never say something is 0% or 100% fake because we don’t have that ground truth. That said, our award-winning models are able to say, with a confidence score of 1-99% — the higher being likely manipulated — which, in turn, is sent to the team using said detection to action as they will. Some use it as one of many signals to make an informed decision manually. Others have chosen to moderate or label accordingly. There are experts who’ve been called to testify on matters like this one, and some of them work on these very models.
As for synthetic content that is undetectable to the naked eye or ear, we are already there.
Thank you! We’ve been working on this since 2021 (and some of us a bit before that), and we’re reminded every day that we are ultimately working something that helps people on the macro and micro level. We want a world free of the malevolent uses of deepfakes for ourselves, our loved ones, and everyone beyond, and feel all should be privy to such protection.
I understand the appeal of hashing-based provenance techniques, though they’ve faced some significant challenges in practice that render them ineffective at best. While many model developers have explored these approaches with good intentions, we’ve seen that they can be easily circumvented or manipulated, particularly by sophisticated bad actors who may not follow voluntary standards.
We recognize that no detection solution is 100% accurate. There will be occasional false positives and negatives. That said, our independently verified an internal testing shows we’ve achieved the lowest error rates currently available for addressing deepfake detection.
I’d respectfully suggest that dismissing AI detection entirely might be premature, especially without hands-on evaluation. If you’re interested, I’d be happy to arrange a test environment where you could evaluate our solution’s performance firsthand and see how it might fit your specific use case.
We see who signs up for Reality Defender and instantly notice traffic patterns and other abnormalities that allow us to see if an account is in violation of terms of service. Also, our free tier is capped at 50 free scans a month which will not allow for said attackers to discern any tangible learnings or tactics they can use to bypass our detection models.
We've actually deployed to several Tier 1 banks and large enterprises already for various use-cases (verification, fraud detection, threat intelligence, etc.). The feedback that we've gotten so far is that our technology is high accuracy and a useful signal.
In terms of how our technology works, our research team has trained multiple detection models to look for specific visual and audio artifacts that the major generative models leave behind. These artifacts aren't perceptible to the human eye / ear, but they are actually very detectable to computer vision and audio models.
Each of these expert models gets combined into an ensemble system that weighs all the individual model outputs to reach a final conclusion.
We've got a rigorous process of collecting data from new generators, benchmarking them, and retraining our models when necessary. Often retrains aren't needed though, since our accuracy seems to transfer well across a given deepfake technique. So even if new diffusion or autoregressive models come out, for example, the artifacts tend to be similar and are still caught by our models.
I will say that our models are most heavily benchmarked on convincing audio/video/image impersonations of humans. While we can return results for items outside that scope, we've tended to focus training and benchmarking on human impersonations since that's typically the most dangerous risk for businesses.
So that's a caveat to keep in mind if you decide to try out our Developer Free Plan.
We are optimizing compute requirements by isolating only the part of the media that requires scanning, so that we can pass along the cost savings to customers. At scale, the costs will low enough to scan all files just like anti-virus.
And we’ve published peer-reviewed research at top AI conferences E.g. CVPR, NeurIPS, ECCV, AAAI, Interspeech which are available at https://www.realitydefender.com/research