A 47-Episode AI Short Drama Was Copied and Only Awarded RMB 20,000 — But One Line in the Judgment Is Worth Far More
Let's start with a case.
A 47-episode AI short drama, one hour in total, was renamed by another company and moved wholesale onto its own video account, where it played with ads inserted along the way.
The court ruled: RMB 20,000 in damages.
My first reaction was, is that it? But then I read one line further down in the judgment, and realized the truly valuable part of this case was never that RMB 20,000.
For the first time, the Jiang'an District People's Court in Wuhan included the token compute cost burned during creation, along with the money spent on commercial AI tools, in the damages calculation.
The cost of AI creation can now be seen by a court. That single line matters far more than the amount awarded.
On what basis was it deemed an "audiovisual work"?
The case could be won only on the premise that the short drama was recognized as a protected "audiovisual work." The judge put it bluntly: this was not a one-click output. The creators were present throughout, checking every stage, and AI was at most a tool.
The judge broke the creative process into several stages: in the early stage, writing the script and drafting the copy; in the middle stage, setting storyboard prompts, selecting character images and scene materials, and screening usable clips from the pile of AI-generated fragments; in the late stage, editing the film, refining details, and syncing the soundtrack and subtitles.
Did you catch it? Everything the judge described was "something a human did." If you want a court to recognize your work, you have to be able to prove that you did these things.
These 3 types of evidence — start retaining them now
Anyone making AI short dramas knows in their bones that running the same prompt one more time means spending money one more time, and that the failed clips pile up beyond counting. Those failed clips are exactly what you need to keep.
Let me translate the stages the judge identified into what you should be retaining in ordinary practice, in three layers:
Layer one, ownership and conception. Don't delete your scripts or draft copy; don't clear your team's records of creative discussions.
Layer two, the generation process. Keep your storyboard prompts one by one; keep records of how you selected materials and screened clips; don't overwrite your tool operation logs or source project files.
Layer three, publication and cost. Keep timestamped screenshots of first publication, the bills for burned tokens, and the receipts for commercial tool licences. These last two were included in a damages award for the first time in this case — they are the easiest to overlook, and also the most concrete.
Why call them the most concrete? When infringement occurs, actual losses and illegal gains are both hard to prove, so the court can only award damages at its discretion. Retaining cost receipts in ordinary times gives the judge one more basis for calculating the money. This judgment says so in black and white.
Winning comes down to evidence retention; losing comes down to patching things up afterwards
For the same category of AI works, two cases in Beijing ended worlds apart.
In the first text-to-image AI case, the plaintiff produced all the prompts, parameters and modification process, and won. In another case involving a cat design, the plaintiff had no process records at all, and tried afterwards to patch things up with "reproduction simulations," which was rejected.
The court's line was simple: whoever asserts a right must explain their creative thinking, what instructions they input, how they selected and modified, and must produce evidence.
The difference is not AI. It is whether you saved it at the time or patched it up afterwards.
One more thing by the way: heavy copying does not only mean paying damages. In Guangzhou, someone illegally recorded over 1,700 AI short dramas, going well past the criminal prosecution threshold, and is already behind bars.
Three things you can do today
First, starting today, create a folder for each project's prompts, drafts and bills, and store them in three slots: "ownership, generation, cost." Don't wait until something happens to go digging.
Second, before deleting chat records or overwriting project files, copy a set to cloud storage and label the date.
Third, on the day you publish, immediately screenshot the publication page with the timestamp, and don't crop the date out of your bill screenshots.
An AI short drama was copied and only awarded RMB 20,000. That number is genuinely unsatisfying. But the line in the judgment about token costs may be the single most useful line for AI creators this year. When it comes to evidence retention, the earlier you start, the cheaper it is.
Next article: the "explicit + implicit" dual labelling that AI-generated content must carry from July 1 — what exactly counts as compliant labelling? Leave a comment below and tell us about the pitfalls you've stepped in.
He Zigang | Intellectual Property Lawyer | Aipunajie · Mili · Najie
This article represents the author's personal views only and does not constitute legal advice. For analysis of a specific case, please contact us.
This article represents only the author's personal views and does not constitute legal advice. For specific case analysis, welcome to contact us.