Two people are kissing on screen, and they’ve even added the sound of their breathing. The whole thing cost 3,000 yuan to make, took twenty days, and runs for thirty episodes.
Watch just this clip and it’s almost impossible to tell whether it was filmed with real actors. It’s trending on Weibo right now, under the headline “AI kiss scenes are even steamier than the real thing.” The show is called *Green Plum by the Inkstone*, made by a female creator who says she spent two months learning the tools and produced thirty episodes in twenty days. An equivalent live-action short-form series is estimated by the industry to cost 500,000 to 1 million yuan. The AI version costs 3,000 to 5,000. That’s a difference of more than a hundredfold.
The output is even more staggering. I looked up the China Network Audiovisual Association’s *Guidelines for Creating Micro-Dramas*: in the first quarter of this year, the industry released about 128,000 micro-dramas, around 122,000 of them AI micro-dramas—more than 95%. In all of last year, the country released 33,000 micro-dramas in total. In one quarter, they produced four times as many as in the entire previous year.
Hengdian was the first to feel the chill. At its peak, 40 or 50 productions would start filming on the same day, with more than 100,000 actors on the books. This year, after the Spring Festival, day rates for supporting actors were cut in half. The market has also spawned a business called “selling your face”: for 100 to 1,500 yuan, actors sell off the permanent rights to their likeness, which is then used to feed digital-human asset libraries.
My take is that this replacement isn’t winning on quality at all. AI short dramas win by bringing the cost per episode down to 3,000 yuan, and by making production infinitely replicable. Once content is no longer worth much, the scarce thing left is whether it’s real.
The same technology was used to swindle 200 million Hong Kong dollars in Hong Kong this January. Arup—the engineering firm behind the Bird’s Nest and the Hong Kong–Zhuhai–Macao Bridge—had a finance employee receive a meeting invitation from someone posing as the CFO at its UK headquarters. Everyone in the video could be seen and heard. The employee transferred the money in fifteen transactions to five local accounts. Police later found that the only real person in the meeting was the one who reported the crime; all the audio and video had been assembled from publicly available YouTube clips. The company’s global information director said that attacks involving forged receipts, phishing, voice fraud, and deepfakes are rising sharply in both number and sophistication.
You see, once faces and voices can both be faked, you can’t trust someone saying “transfer the money immediately” in a video meeting, either. The crypto world’s mantra, “Don’t trust, verify,” used to sound like a slogan. Now it looks like a survival skill. Satoshi Nakamoto once used a digital signature to prove he was himself, without showing his face. Valuable on-chain actions are all tied to signatures, not faces. Behind those 122,000 AI micro-dramas is a massive amount of GPU inference. The demand for rendering is real, and distributed computing networks are targeting exactly that.
The weak spots are showing, too. In the second quarter, 239,000 micro-dramas were released nationwide: 86,000 live-action and 153,000 AI-made. AI’s share fell from 95% to just over 60%. The Spring Festival release period was even more telling: the number of live-action dramas released was just one-fiftieth the number of AI dramas, yet their total views were twenty-five times higher. Audiences may marvel at AI, but their fingers are still voting for real people.
AI’s share had already fallen by the second quarter, and the guidelines only cover the first quarter. I’m not sure it can hold at 90% for the full year.
A kiss scene made for 3,000 yuan and a lesson bought for 200 million Hong Kong dollars are two sides of the same story. Technology has dismantled the default setting that says “seeing is believing.” From now on, only what can be verified will remain.
$BTC $ETH $RNDR #AI短剧 #深度伪造 #微短剧 #奥雅纳 #币安广场
Watch just this clip and it’s almost impossible to tell whether it was filmed with real actors. It’s trending on Weibo right now, under the headline “AI kiss scenes are even steamier than the real thing.” The show is called *Green Plum by the Inkstone*, made by a female creator who says she spent two months learning the tools and produced thirty episodes in twenty days. An equivalent live-action short-form series is estimated by the industry to cost 500,000 to 1 million yuan. The AI version costs 3,000 to 5,000. That’s a difference of more than a hundredfold.
The output is even more staggering. I looked up the China Network Audiovisual Association’s *Guidelines for Creating Micro-Dramas*: in the first quarter of this year, the industry released about 128,000 micro-dramas, around 122,000 of them AI micro-dramas—more than 95%. In all of last year, the country released 33,000 micro-dramas in total. In one quarter, they produced four times as many as in the entire previous year.
Hengdian was the first to feel the chill. At its peak, 40 or 50 productions would start filming on the same day, with more than 100,000 actors on the books. This year, after the Spring Festival, day rates for supporting actors were cut in half. The market has also spawned a business called “selling your face”: for 100 to 1,500 yuan, actors sell off the permanent rights to their likeness, which is then used to feed digital-human asset libraries.
My take is that this replacement isn’t winning on quality at all. AI short dramas win by bringing the cost per episode down to 3,000 yuan, and by making production infinitely replicable. Once content is no longer worth much, the scarce thing left is whether it’s real.
The same technology was used to swindle 200 million Hong Kong dollars in Hong Kong this January. Arup—the engineering firm behind the Bird’s Nest and the Hong Kong–Zhuhai–Macao Bridge—had a finance employee receive a meeting invitation from someone posing as the CFO at its UK headquarters. Everyone in the video could be seen and heard. The employee transferred the money in fifteen transactions to five local accounts. Police later found that the only real person in the meeting was the one who reported the crime; all the audio and video had been assembled from publicly available YouTube clips. The company’s global information director said that attacks involving forged receipts, phishing, voice fraud, and deepfakes are rising sharply in both number and sophistication.
You see, once faces and voices can both be faked, you can’t trust someone saying “transfer the money immediately” in a video meeting, either. The crypto world’s mantra, “Don’t trust, verify,” used to sound like a slogan. Now it looks like a survival skill. Satoshi Nakamoto once used a digital signature to prove he was himself, without showing his face. Valuable on-chain actions are all tied to signatures, not faces. Behind those 122,000 AI micro-dramas is a massive amount of GPU inference. The demand for rendering is real, and distributed computing networks are targeting exactly that.
The weak spots are showing, too. In the second quarter, 239,000 micro-dramas were released nationwide: 86,000 live-action and 153,000 AI-made. AI’s share fell from 95% to just over 60%. The Spring Festival release period was even more telling: the number of live-action dramas released was just one-fiftieth the number of AI dramas, yet their total views were twenty-five times higher. Audiences may marvel at AI, but their fingers are still voting for real people.
AI’s share had already fallen by the second quarter, and the guidelines only cover the first quarter. I’m not sure it can hold at 90% for the full year.
A kiss scene made for 3,000 yuan and a lesson bought for 200 million Hong Kong dollars are two sides of the same story. Technology has dismantled the default setting that says “seeing is believing.” From now on, only what can be verified will remain.
$BTC $ETH $RNDR #AI短剧 #深度伪造 #微短剧 #奥雅纳 #币安广场