First, this isn't an ad. I recorded this video quite suddenly because ByteDance just released a model called Seedance 2.0. My normal attitude toward AI models is: oh, another one released, then a lot of impressive concepts get hyped, then the actual experience isn't great, then I move on. But this model, when I first saw the video, I thought: oh, this can do that too — look at this, for example. The key is its precision, right? This is really impressive. Then I saw people online starting to produce content, like this and this. I'm not saying it's perfect, but you can clearly feel it's much stronger than previous AI video, right? At that point I felt this model isn't ordinary. Then our own company could start producing content, I tried it myself, and my feeling is: you really have to try it yourself. If video-generation models are already at this level at the start of 2026, I feel the traditional film workflow has truly entered a countdown to being washed away by an AI tsunami. I'm not fear-mongering. Let's look together at what video-generation models are today. Past video-generation models were already quite good and looked convincing, but there were points where you could break them: large-scale camera movement, shot continuity, and audio-visual matching. Let's look at Seedance 2.0 across these three pain points. First, large-scale movement: we asked our company's AI director to write a text and an image without any other information, and this is our result. Isn't this stunning? This kind of camera movement is now something AI can generate. What, you want an even crazier camera move? No problem, we change the prompt and look again. I think that's enough said. In the past we identified AI video by whether the camera movement looked fake; this method has basically become invalid. Now the second point, editing. Previous models could generate cuts, but old cuts were like this: they cut for the sake of cutting, at most a close-up then a wide shot, and the camera movement itself didn't connect. Let's look at the most staggering part of this model: the editing design of a video generated from one sentence. I don't know if you noticed, it's not only close-ups, there's a clear switching of angles, there's a director's intent, and the characters don't deform. Maybe that sounds unclear, so let's use a top-down view to simulate the change of subjects. You see it cuts here, here, here — like a real director constantly changing the camera's position and angle so the audience understands the key content better. This is a director's thinking. There's a focus point. In the past I thought AI would generate a clip and an editor would cut them together into a great work. But now we have to realize: once such models mature, editors won't be needed. There's no concept of filtering or discarded shots; everything is usable. As long as the generated cuts are good enough, why do I need editing? I used to really believe editing is an emotional expression — insert a few frames and the mood changes, so a person must do it. But now we have to face this question: how are you sure the frames you insert are better than the frames AI inserts? Especially since it can give you countless versions, very fast. Now the last point, audio-visual matching. A video without sound loses a lot. The biggest problem with AI before was that the dubbed sound was fake; especially human speech plus ambient sound plus music together felt unnatural. Let's look at this model's result — pretty good, right? You should know that doing SFX and mixing used to require sticking things on one by one by hand, taking a day or more; now it's done in one click. But there are quite a few excellent sound models in the industry now, so this won't feel stunning. But I want to mention something we accidentally discovered that's a bit terrifying: I found a point — as long as I upload my own face to Seedance 2.0, this AI automatically speaks with my voice. I'm not sure you understood what I said — without any prompt, any words, any information, I didn't even give my voice file, I just uploaded my face — this isn't an ad — and the AI actually knew that this face's voice is mine. Let me demonstrate: this is the result we generated with one sentence plus my photo. My normal attitude to AI models is oh oh another release... but this model's demo video made me think: oh, isn't this terrifying? And the most terrifying part isn't only my voice — the photo we uploaded was like this, only the front of the building, but look at the camera movement, it rotates to the other side of the building. What does the other side of our building look like in reality? Like this. Isn't that terrifying? It knows what's behind me even though I didn't tell it. This basically confirms one thing: Seedance 2.0 has trained heavily on our company's videos. You'll ask, didn't they ask for your copyright? Simply put, the platform's user agreement, I don't know, maybe a similar licensing clause is hidden inside, but personally I haven't received money or been contacted for authorization. Maybe they're legally compliant, but it's a bit terrifying, right? Even more terrifying, we tested other creators, like He Tongxue, and you'll find the character consistency is super high. The sand we normally see is formed from rock through long weathering, a gift of time to us… well, maybe the audio quality of He Tongxue's video still has room to improve. But we found that Seedance this time chose to use my voice for him, and many others inexplicably got my voice too. It's a bit funny, but after laughing we have to get serious again. I truly feel we all need to wake up now, because we count as a case: among the earliest to upload all information and data, especially image and audio data, to the cloud, and also me as an individual, especially HD content. You can now see: if a person's data all enters the AI dataset, what happens? It can 100% simulate any form and voice of yours. Could your family tell it apart? I think by now you understand why we made this video. This isn't a small technical innovation, this is a tsunami that will sweep away all the previous workflows and accumulated knowledge of the industry — that's how I'd describe it. Before the end I want to give another case: a while ago we put out a video — AI can replace me, so what's my meaning? In that video we told how we helped a fan with a tumor make an MV, and spent two years on effects, but in the end we found that the harder we tried, the more we couldn't beat the speed of AI's progress. What I learned before is losing value. In 2022 and 2023, to handcraft a shot of a train flying into the sky took dozens to hundreds of hours. In 2023's AI, minutes could produce such a result, not very good. In 2025, two years later, by the mid-term, AI could produce such a result, already very strong, right? So what about today's Seedance 2.0? We tried it too, this is its current result. So enough talk, I truly think you should go try this AI model yourself to better understand how scary it is. It's not yet the AI that will truly change the video industry — I say that from the bottom of my heart — but the next AI, the next version, might be it. So is this good or bad for humanity?