Ve Working Towards
Pangram extending from text detection into visual media is a logical expansion, and the hard problems the author names are the right ones: computational photography genuinely blurs the 'real' category, a nuance most detectors ignore. In our view the heatmap is the strongest differentiator — localization turns a fuzzy confidence score into auditable evidence, which is what moderation and forensics teams actually need.… more
The edit-versus-generated distinction matters too, since a touched-up photo and a pure deepfake deserve very different treatment. The catch is the adversarial treadmill: every fresh generator from an obscure lab breaks detectors until retrained. Pangram also enters a field with entrenched players like Hive and Reality Defender. If accuracy survives independent testing, this is a serious entry — the provenance wars are only starting.
We've been working towards AI image and video detection for months, and now it's finally here.
There were a lot of challenges along the way. There are way more AI image/video models than frontier LLMs, and their outputs are more diverse. Most phone cameras, including iPhones, do some level of AI postprocessing to add detail to real photos.
We ha…
Source:
X — Max Spero
Discussion (0)