A new AI system from ByteDance, OmniHuman-1, generates shockingly convincing deepfake videos.
  • 𝕸𝖔𝖘𝖘@infosec.pubEnglish
    13·
    12 hours ago

    Trained on 19,000 hours of video content from undisclosed sources

    I’m gathering their “undisclosed sources” are social media (YouTube, Instagram, TikTok, Facebook, etc.) and copyrighted content (DVDs, BRDs, etc.), all without permission or legal rights. I wonder when pirating becomes just another brick in the wall…

    • Em Adespoton@lemmy.ca
      6·
      12 hours ago

      Well, now they have the ability to create new derivative content on the fly… next up is China mass producing movies and TV shows for all regions, with the content subtly different for each country or target demographic.

  • melp@beehaw.orgEnglish
    4·
    10 hours ago

    “According to the ByteDance researchers, OmniHuman-1 only needs a single reference image and audio, like speech or vocals, to generate a clip of an arbitrary length.”

    I am SO glad I never did a youtue channel, a vine, an IG, or really anything. This is creepy as fuck.

  • Arghblarg@lemmy.ca
    18·
    14 hours ago

    This is scary indeed. We may someday soon need something like an active tattoo on our face, or a badge on clothing, with a pattern that changes each second based on a private/public key pair, so videos can’t be easily faked of our own likeness with a valid visual signature.

    That could actually work – a QR code that updates at regular intervals, encoding an ever-changing signature. It could be validated to certify the video of a person was genuine.

    Of course that would also mean any authenticated video can never be truly anonymous :(

    • drspod@lemmy.ml
      8·
      14 hours ago

      Of course that would also mean any authenticated video can never be truly anonymous :(

      With modern face recognition, having your face in a video already deanonymizes you so there is no further deanonymization caused by authenticating your video cryptographically.

      If you want anonymity then don’t include your face (or signature) in the video.

      If you want attestation that a video is real and not generated then anyone who witnessed the scene depicted can sign the video to attest to it. Then we would need a web of trust to determine whose signatures are trustworthy.

      • Arghblarg@lemmy.ca
        3·
        13 hours ago

        With modern face recognition, … If you want anonymity then don’t include your face (or signature) in the video.

        Duh, good point :)

        Web of trust – it’s always been so hard to make easy enough to use for the non-technical public, sadly… but yeah that might be the only/best way to really give attestation.

    • kbal@fedia.io
      4·
      14 hours ago

      a QR code that updates at regular intervals, encoding an ever-changing signature

      That’s brilliant. Any sci-fi authors in the crowd? The private key would be used to hash and sign a record of body movements including all one’s crazy hand gestures and every word spoken. Location could optionally be encoded as well. Some kind of algorithm would be devised so that some reasonable loss in fidelity of any video recording would still result in a valid signature. It wouldn’t be technically all that useful to display the signature as a QR code shown on a badge, but it’d be a fashionable thing to do and anyone who didn’t would be seen as slightly suspicious.

      Key infrastructure would be tricky, but anonymous and pseudonymous keypairs could certainly be allowed for if we go with the assumption that instant biometric identification of everyone isn’t quite feasible for whatever reason. Maybe it’s just banned, punishable by exile to the orbital asteroid mining colonies.

      All we need is for everyone (except the underclass) to get neural implants that record their every movement.

      • Arghblarg@lemmy.ca
        6·
        13 hours ago

        exile to the orbital asteroid mining colonies.

        Can we start with Sam Altman please? Hah.

        I was amazed at first with ChatGPT, outpainting, and the early stuff; it was fun making ‘paintings’ and playing with other imagery, but the main uses are taking such a dark turn I really think we’re going to regret this technology’s existence.

  • jarfil@beehaw.org
    4·
    12 hours ago

    Hands, hair, camera angles, and voice sync, still have some way to go… but they do seem like they’re slowly climbing out of the Uncanny Valley.

    Interesting times.

  • kbal@fedia.io
    4·
    15 hours ago

    Albert Einstein saying “science does not answer ‘why’ questions” isn’t realistic at all.