The AI video game just hit god-mode. ByteDance, the company that brought us TikTok and CapCut, dropped Seedance 2.0 last week, and it\u2019s straight-up chaos in the best way. This beast of a tool lets you mash text, pics, clips, and audio into hyper-real videos that look like they came from a Hollywood blockbuster. Natural moves, lip-sync that slaps, lighting on point\u2014no more creepy Uncanny Valley vibes.<\/p>\n
![\"RADII](\"https:\/\/radii.co\/wp-content\/uploads\/2026\/02\/radii-seedance-2.0-ai-gone-wild-01.jpg\")
Image via CapCut.<\/p>\n
In just a few days, creators flooded the feeds with deepfake-style clips starring a range of celebrities. Think Kanye West crooning a classic Chinese song<\/a> in an ancient palace (after storming into a room with ex-wife Kim Kardashian surrounded by clones of Travis Scott and iShowSpeed<\/a>). Or Justin Bieber in front of the Hong Kong skyline, and then Jet Li versus Jackie Chan in an old movie setting<\/a>. Or Stephen Chow<\/a> goofing around with NBA star Kawhi Leonard<\/a> in a scene from Fight Back to School 2. Other generated clips included Disney\u2019s Spider-Man, Iron Man, and Grogu aka \u201cBaby Yoda.\u201d<\/p>\n Irish film-maker Ruair\u00ed Robinson, director of The Last Days on Mars, posted a 15-second clip featuring Brad Pitt and Tom Cruise throwing punches at each other in an intense rooftop scene. The fabric moves right. Their hair physics track naturally. Their facial expressions sync perfectly with the dialogue. The lighting matches the scene. It\u2019s not deepfake jank anymore\u2014it\u2019s Hollywood-level output. All this with just \u201ca 2 line prompt\u201d and likely under 1 minute of rendering. Not to mention that Seedance 2.0 is currently free to try, and ByteDance\u2019s full AI video suite is accessible starting from $18 per month. Rhett Reese, co-writer of Deadpool & Wolverine, shared Robinson\u2019s clip on X and said, \u201cI hate to say it. It\u2019s likely over for us.\u201d<\/p>\n It helps to understand how easy it is to operate Seedance 2.0. This tool is multimodal magic\u2014just feed it up to 12 assets, which include up to 9 images, up to 3 short videos (no longer than 15 seconds), and up to 3 audio files (also under 15 seconds), plus text prompts to let the AI know how to use those assets as hard references. Want a vid of Drake performing in a neon dystopia? Upload his pic, a beat drop, and describe the vibe. Boom: physics-real motion, fabric that flows right, faces that emote perfectly. It syncs audio to visuals \u2014 lip-sync dialogue, sound effects that match the action, even choreographs to your uploaded track.<\/p>\n While we mere mortals will get a good laugh out of Seedance 2.0, pros will likely eat this up given the cost-savings for such quality, consistency and speed. TikTok<\/a>-ready 9:16 or YouTube 16:9? Done. Extend scenes, swap characters, tweak frames\u2014all post-gen. ByteDance calls it the \u201cmost comprehensive multimodal content reference\u201d via their Seed site, topping benchmarks in text-to-video, image-to-video, and more.<\/p>\n Hollywood is, understandably, very concerned about this upgraded GenAI<\/a> tool. Within days of Seedance 2.0\u2019s launch, Disney and Paramount fired off cease-and-desist letters to ByteDance, alleging massive copyright and right-of-publicity infringement. In a statement, the Motion Picture Association (MPA) also accused ByteDance of \u201cunauthorised use of US copyrighted works on a massive scale.\u201d<\/p>\n Their argument? These AI videos are training on protected content. Those Jet Li deepfakes? Probably trained on his actual filmography. The Jackie Chan vids? His movies, his likenesses, his signature moves. The Kanye West deepfakes? His image, his voice, his persona\u2014all legally protected intellectual property. You can\u2019t just run an AI on that content and then generate infinite variations without compensation.<\/p>\n But it gets messier. The deepfakes of Tom Cruise and Brad Pitt hit a different legal nerve because both actors have massive franchises attached to them. Cruise is literally Mission: Impossible, a crown jewel for Paramount. Pitt has a massive IP tied to his image through various studios. When you generate convincing videos of these guys doing things they never did, wearing things they never wore, and potentially saying things they never said, you\u2019re messing with right of publicity\u2014the legal right to control how your image and likeness are used for commercial purposes.<\/p>\n The wildcard? No one knows exactly what Seedance 2.0 was trained on. ByteDance hasn\u2019t fully disclosed the training data. But given that it can replicate specific celebrity likenesses and mannerisms so perfectly, studios are assuming the model ingested massive amounts of footage, photoshoots, and audio recordings from their libraries and the broader internet.<\/p>\n Disney\u2019s particularly pressed because they\u2019ve got Marvel, Star Wars, and decades of legacy content. If Seedance 2.0 can generate convincing deepfakes of characters tied to that IP, it\u2019s a threat to their entire business model. Having said that, back in December, OpenAI signed a deal that allows it to feature Disney characters on its Sora video generator, so it\u2019s just a matter of money.<\/p>\n What\u2019s undeniable is the fact that advancements like Seedance 2.0 democratize content creation in ways that scare massive corporations. Anyone with a laptop or even a phone can now generate Hollywood-quality videos. No crew. No budget. No gatekeeping. You get unlimited creative variations without reshoots. That\u2019s genuinely world-changing for independent filmmakers, TikTok creators, small businesses, and literally anyone who wants to make video content without bleeding money.<\/p>\n But the flip side? Consent and ethics are getting obliterated. These celebrities never agreed to have their likenesses used this way. Kanye didn\u2019t authorize his image to be used in deepfakes (well, probably\u2014knowing Kanye, he might actually be here for the chaos, but that\u2019s besides the point). The broader question is: should AI be allowed to replicate people\u2019s identities without their permission?<\/p>\n Legally, studios are fighting for \u201cno.\u201d Creatively, the internet is fighting for \u201cwho cares, this is fire.\u201d And technically, ByteDance<\/a> isn\u2019t stopping it. In fact, the tech isn\u2019t going away. Other companies are building their own versions. The genie is out of the bottle, and you can\u2019t shove it back.<\/p>\n The future is probably some hybrid: celebrity consent protocols where creators have to flag if they\u2019re using a public figure, licensing agreements that compensate studios and talent, and clear legal frameworks for what constitutes fair use in AI<\/a> generation.<\/p>\n For now? Enjoy the chaos. Download those Jet Li deepfakes. Duet the Kanye<\/a> videos. Remix the Justin Bieber clips. Because this might be a very brief window where we can create anything before the lawyers permanently close the door.<\/p>\n Cover image via YouTube.<\/p>\n<\/p><\/div>\n![\"RADII](\"https:\/\/radii.co\/wp-content\/uploads\/2026\/02\/radii-seedance-2.0-ai-gone-wild-02.jpg\")
Yup, that\u2019s Mixue\u2019s Snowman King battling it out against Luckin Coffee and Starbucks robots. Image via hk01.<\/p>\n