as isFake is a multimodal detector, you can check text, images, videos and audio on our website. all in one place 😉
every model has its own features. video and audio detectors highlight the most suspicious fragments of the content, while image detector uses the heatmap to show the most AI-generated areas of the picture
text detector also highlights separate phrases that are, most likely, AI
we also have articles on our website that describe how each detector works and explain, how to read detection results and reports
speaking of a @ feature, we haven't introduced it yet. but we actually plan to do it in the future, so stay tuned! 😉
thanks a lot for your interest in isFake! have you tried it already?
This is a completely AI-generated video.
You can notice that some fragments are highlighted in red, while others are not highlighted at all. That doesn't mean they are real, that just means those particular frames lack enough movement for a detector to be able to spot AI in them.
When the frame is very stable (especially like the last one), it's getting hard to analyze it properly, as the detector needs movement to notice that something's off. Still, 67% confidence score is enough to be conserned about content authenticity
This is a completely AI-generated video.
You can notice that some fragments are highlighted in red, while others are not highlighted at all. That doesn't mean they are real, that just means those particular frames lack enough movement for a detector to be able to spot AI in them.
When the frame is very stable (especially like the last one), it's getting hard to analyze it properly, as the detector needs movement to notice that something's off. Still, 67% confidence score is enough to be conserned about content authenticity
When we train our detection models we feed them with a lot of different content–both real and AI. And we get the new fresh models that are able to identify the most modern AI-generated videos, images, audios and text, but…
BUT there’s one thing that depends only on you.
This is your own viewing experience and perception of AI.
You see, creating a detector is practically useless, if a user doesn't understand when they should use it. Sorry for the repetitions, but that’s actually what we mean 😄
If someone sees a video and doesn’t get the ‘uncanny valley’ effect and that brief thought ‘Oh, that might be AI’, then they would not even get the brief thought ‘Oh, maybe I should check it with isFake’
And that’s exactly the thing that depends on you. Seeing, exploring, trying out different genAI models to get the idea of what they are capable of.
And don’t worry, we know how to handle this. On our website we have an ‘Explore’ page where we post recent AI cases. That’s like Pinterest, but for AI–scroll, discover and share 😉
And whenever you see something weird on your feed, we are always there for you to make sure–is it real or AI?
@keza_250 that's so funny that even in the gibberish airline name there's 'AI' part
completely fake video, checking through isFake AI detector as always 😉
@Architect_ONI even the wisest Indian monks can't climb vertical mountains like this. yet.
this video is AI-generated, and we've checked with isFake AI video detector. you can try it yourself as well, especially before sharing some 'shocking' videos 😏
that's a completely AI-generated clip
63% confidence score from our AI detection model should not consern you. the thing is that the audio was not detected as AI, while the majority of frames are highighted as AI with strong confidence (more than 80%). this contrast creates those 63%.
but the report reveals all the details and it becomes pretty clear: the video is fake ;)
that's AI-generated
but why are frames in the middle of the video not highlighted? let us explain
when checking any video, AI detector analyzes the frames and compares how the neighboring frames change. to analyze these changes properly, it needs those frames to have movement. so, basically, when there's not enough movement in the frame, it's getting hard for a detector to compare it with the previous one and with the next one.
here's that exact case. you can notice that the 'not-highlighted' part of the video just lacks movement in the frames, it's almost stable
but even with that, 69% is a very strong sign of AI