Hollywood IP crackdown: Sora shutdown/Disney + Seedance 2.0 guardrails + Veo 3.1 + deepfake detection surge + labeling
Key Questions
What is Seedance 2.0 and where is it available?
Seedance 2.0 is a viral AI video model now available on Runway and CapCut, with tutorials on SoraVideo.art for beginners. It continues its takeover in the AI video space, offering step-by-step guides for testing prompts.
What deepfake incidents have escalated recently?
Recent deepfakes include a fake disease hoax called Bixonimania spread by AI, AI-generated war misinformation images, and pornographic images of students created by teachers. Cybercrime tools like Jinkusu target crypto and bank KYC systems using deepfakes.
What new laws address AI deepfakes and disclosure?
Washington's HB1170 requires disclosure, watermarks, and sue-rights for AI content. Provenance Marker Law mandates global AI fingerprints like C2PA and SynthID for content labeling.
How is Apple addressing AI safety?
Apple's Quiet AI Labs develop tools like SafetyPairs for VLM safety and image safety rating to flag unsafe images. Their research also explores AI-assisted UI prototyping.
What are some new deepfake detection tools?
Tools include Resemble DETECT-3B, CLIDE, ProofMark, Moonbounce, and Reality Defender. Insurers like Verisk use 50% homegrown detection, but studies warn many methods can be easily fooled.
What is Veo 3.1?
Veo 3.1 is a new AI video model mentioned alongside Seedance 2.0 guardrails. It contributes to the surge in video generation amid IP crackdowns.
What watermark and detection standards are emerging?
ASU Yang proposes watermark standards with detection traces and T2I unlearning via RACE/EraseFlow. Edinburgh highlights vulnerabilities in current systems.
Why is there a Hollywood IP crackdown?
It involves Sora shutdowns, Disney protections, and guardrails on models like Seedance 2.0. This responds to deepfake surges in misinfo, hoaxes, and IP violations.
Deepfake escalation with fake disease hoax, war misinfo (Hive/Grok), insurers 50% homegrown (Verisk), Jinkusu KYC, Farid talk, Edinburgh vulns, WA HB1170 disclosure/watermarks/sue-rights, Provenance Law C2PA/SynthID; now ASU Yang watermark standards/detection traces/T2I unlearning (RACE/EraseFlow), Apple SafetyPairs VLM safety. Seedance 2.0 CapCut/SoraVideo.art tuts; Veo 3.1; Resemble DETECT-3B/CLIDE/ProofMark; Moonbounce/Reality Defender.