Key takeaways
- At Sora's debut roughly two years ago, generating one second of video cost about US$0.50. Today the consumer API pricing of Chinese models — Kling, Jimeng, Vidu, Wan 3.0 — has pushed the per-second price below RMB 0.01.
- That is a 99% decline in about 24 months.
- In a conventional video project, rough cutting plus subtitles and voiceover account for 60–70% of hours — precisely the parts AI handles best.
- New roles with published salary ranges: AI prompt engineer (video), RMB 15,000–25,000 a month; video QC specialist, RMB 8,000–12,000 a month; and copyright clearance, already a function inside ByteDance, Alibaba and NetEase.
The editor who lost half an order to a junior
Have you seen the advertisements claiming "one mao per second" AI video?
An eight-year veteran editor in Hangzhou — call him Lao Li — recently lost half an order to a third-year intern who knows AI. The intern produces in one evening what Lao Li produces in a week.
Is he worried? Yes. But what he realised is more interesting than the worry.
AI is not taking the editor's job. It is re-cutting it.
The first thing to state plainly: AI is not here to steal editing work. It is here to redivide it.
A conventional video project runs writer — storyboard — shoot — rough cut — fine cut — colour — voiceover — subtitles. Of those eight stages, rough cut, subtitles and voiceover consume 60 to 70% of the hours, and those are exactly what AI is best at.
What AI can do now:
- generate a rough cut automatically, including complete shots from a script
- generate subtitles, switching live between ten languages
- generate voiceover, with voice cloning and lip sync
- generate storyboards from copy and reference images
What is left for the editor is the more valuable part: creative direction, control of pacing, brand-grade colour, and talking to the client.
AI cannot do those things yet.
The new job map
Once rough cutting is absorbed, the industry does not die. It grows new roles.
AI prompt engineer, video track. Not merely writing prompts, but understanding the language of storyboard, shot and narrative well enough to turn a script into instructions a model can follow. One Hangzhou AIGC studio is offering RMB 15,000–25,000 a month.
Video QC specialist. AI-generated footage needs a human to remove ghosting, deformation and lip-sync errors. It is a completely new trade with a low entry barrier — sensitivity to images is enough — at RMB 8,000–12,000 a month.
Copyright clearance specialist. The largest latent risk in AI video is using someone's face or music without noticing. AIGC companies now need people to check training data, compare similarity and handle complaints. The role already exists inside ByteDance, Alibaba and NetEase.
AI competes on throughput. Humans compete on throughput plus taste.
Here is the analogy that fits: AI has driven the cost of making bricks to zero, but the ability to build a castle is still human.
For the past decade, "can edit" was enough to earn a living from the craft. After AI, what is genuinely scarce in the market is no longer technique. It is taste and judgement.
The owner of a Hangzhou MCN put it directly: they do not need editors who can cut. They need people who understand direction and story, who can say definitively whether a shot stays or goes. That kind of judgement has a long road ahead of it before AI reaches it.
What to do now
If you are an editor: learn one thing this week — hand the rough cut to AI and keep fine cut, colour and client liaison for yourself. Arm yourself with AI first, then let it do the work.
If you want into the AIGC industry: prioritise prompt engineering over editing software. AI tools will keep changing; the language you use to speak to AI will hold its value.
If you are starting a business: look at copyright clearance and video QC. Fewer than 5,000 people in China can be hired into these roles today against a gap estimated at 50,000 to 100,000. A training-plus-placement model has a real opening here.
AI will not replace editors. Editors who use AI will replace editors who do not.
That is not a race to the bottom. It is a level change. From today the job is no longer a technician's; it is a maker's. You have to be able to use AI — and more than that, you have to be able to tell AI how.
Honest limitations
The "one cent per second" figure reflects entry-level API pricing at low tiers and excludes the cost of iteration, failed takes and human finishing, which dominate real budgets. Salary figures are job-posting ranges, not guaranteed compensation. New-role headcount gaps are industry estimates.
Sources: published API pricing for Chinese video models; job postings and studio interviews (August 2026). Information only.