"你告我偷你的东西训练AI,那你敢不敢告诉观众,你自己拍电影的时候用了多少AI?"
这大概就是Midjourney最新法律动作的潜台词。7月5日消息,正在被迪士尼、环球影业、华纳兄弟三家好莱坞巨头联合起诉的AI图像生成公司Midjourney,向法院提交了一份法律申请:要求强制三家制片厂披露其自身使用AI的全部细节。AI版权战打到第二年,终于出现了最具戏剧性的反转。
版权战的罗生门
这场官司的起因大家都很熟悉。去年,迪士尼和环球率先起诉Midjourney,指控其AI模型在未经授权的情况下抓取了大量版权图像进行训练,包括巴特·辛普森、达斯·维达等制片厂旗下的经典角色,用户可以通过prompt直接生成这些角色的图像,构成大规模侵权。几个月后华纳兄弟也加入诉讼。
Midjourney的辩护思路一直是"合理使用"(Fair Use)——训练AI模型抓取互联网上的公开图像,属于 transformative use(转换性使用),就像人类艺术家学习前辈作品一样,不构成侵权。但这个辩护在之前的几起AI版权案中胜算不大,法官倾向于认为大规模抓取版权内容训练商业模型超出了合理使用的边界。
但这次Midjourney换了个打法。它没有反复争辩"我没侵权",而是直接向法官申请:让三家制片厂把他们自己用AI的记录也交出来。具体包括:所有在Midjourney平台上使用过的prompt和对应的生成图像、制片厂内部研发的图像生成AI工具、用AI做影视分镜和创意构思的全部文档。Midjourney的逻辑很简单:如果你们自己也在偷偷用AI生成内容、也在用无授权版权素材训练内部工具,那你们凭什么告我?
制片厂的软肋
为什么Midjourney这招叫"打蛇打七寸"?因为好莱坞大制片厂确实在大规模使用AI,但他们不敢让观众知道。
之前一名法官已经裁定,制片厂必须提供其"面向消费者"的视频和图片中使用AI的信息。但Midjourney认为这个限制太窄了——制片厂只需要挑那些对他们有利的材料提交,而把真正能证明行业惯例的内部文件藏起来。Midjourney想要的是全部:前期分镜用没用AI?概念设计用没用AI?海报修图用没用AI?预告片剪辑用没用AI?
这恰恰是好莱坞最不想公开的部分。2023年好莱坞演员和编剧大罢工,核心诉求之一就是限制AI在影视制作中的使用,保护人类从业者的工作。编剧工会和演员工会好不容易和制片厂达成了AI使用的透明度协议,但制片厂在实际执行中一直是"能不说就不说"。观众如果知道某部斥资2亿美元的大片里,一半的概念图和分镜是AI画的、甚至部分背景角色是AI生成的,会是什么反应?
更有意思的是Midjourney提出的一个论点:如果制片厂内部也普遍存在下载无授权版权素材训练AI的行业惯例,那说明这种行为是行业默认的合理做法,不能只拿Midjourney开刀。这个逻辑不一定能赢官司,但足够恶心对手——你不是站在道德高地上吗?先把你自己的脏衣服晾出来看看。
这场官司的最终结果可能要等好几年,但Midjourney这招"反诉+证据开示"已经改变了游戏规则。AI版权不再是"大公司告小创业公司"的单向叙事,变成了"大家其实都在用,凭什么只说我"的互撕。对整个AI内容行业来说,这未必是坏事——当原告和被告都有把柄在对方手里,最终最有可能的结果不是谁赢谁输,而是坐下来谈一个各方都能接受的授权机制。毕竟,真把AI工具全禁了,好莱坞自己的制作成本也要涨一大截。
“You sue me for stealing your stuff to train AI — but do you dare tell audiences how much AI you use when making your own movies?”
That's roughly the subtext of Midjourney's latest legal maneuver. On July 5, AI image generation company Midjourney — currently being jointly sued by three Hollywood giants: Disney, Universal Pictures, and Warner Bros. — filed a legal motion asking the court to compel all three studios to disclose full details of their own AI usage. In year two of the AI copyright war, the most dramatic reversal has finally arrived.
The Copyright Rashomon
The origins of this lawsuit are familiar. Last year, Disney and Universal took the lead in suing Midjourney, alleging its AI models scraped massive amounts of copyrighted images — including studio-owned classic characters like Bart Simpson and Darth Vader — without authorization, and that users could generate images of these characters directly via prompts, constituting mass infringement. Warner Bros. joined the suit months later.
Midjourney's defense has consistently been “fair use” — scraping public images from the internet to train AI models constitutes transformative use, analogous to human artists studying the works of predecessors, and doesn't constitute infringement. But this defense hasn't fared well in previous AI copyright cases; judges have tended to find that mass scraping of copyrighted content to train commercial models exceeds fair use boundaries.
But this time Midjourney changed tactics. Instead of repeatedly arguing “we didn't infringe,” it directly petitioned the judge: make the three studios hand over their own AI usage records too. Specifically: all prompts used on the Midjourney platform and corresponding generated images; internally developed image generation AI tools at the studios; all documentation of AI used for storyboarding and creative ideation. Midjourney's logic is simple: if you're also secretly using AI to generate content, and also using unauthorized copyrighted material to train internal tools, then on what grounds are you suing me?
The Studios' Weak Spot
Why does this move hit the studios where it hurts? Because Hollywood's major studios do use AI at scale — but they dare not let audiences know.
A judge had previously ruled that studios must provide information on AI usage in their “consumer-facing” videos and images. But Midjourney argues this restriction is too narrow — studios can cherry-pick favorable materials to submit while hiding internal documents that would demonstrate actual industry practice. Midjourney wants it all: was AI used for pre-production storyboards? For concept design? For poster retouching? For trailer editing?
This is precisely the part Hollywood least wants to publicize. The 2023 Hollywood actors' and writers' strike had limiting AI in film/TV production — protecting human practitioners' jobs — as one of its core demands. After WGA and SAG-AFTRA finally negotiated AI transparency agreements with the studios, the studios' actual enforcement has consistently been “don't ask, don't tell.” How would audiences react if they knew that half the concept art and storyboards in a $200 million blockbuster were AI-generated, or that some background characters were AI-created?
Even more interesting is an argument Midjourney raises: if downloading unauthorized copyrighted material to train AI is also a widespread internal practice at the studios, that suggests it's an industry-accepted reasonable practice — you can't single out Midjourney. This logic may not win the case, but it's enough to make opponents uncomfortable — aren't you standing on the moral high ground? Let's air your dirty laundry first.
The final outcome of this lawsuit may take years, but Midjourney's “countersuit + discovery” gambit has already changed the rules of the game. AI copyright is no longer a one-directional narrative of “big companies suing small startups”; it's become a mutual mudfight of “everyone's actually using it, why are you only calling me out?” For the AI content industry overall, this may not be a bad thing — when plaintiffs and defendants both have dirt on each other, the most likely outcome isn't a winner-takes-all verdict but sitting down to negotiate a licensing mechanism acceptable to all parties. After all, if AI tools were truly banned entirely, Hollywood's own production costs would skyrocket.
好莱坞不是反对AI,他们只是反对别人用AI赚他们的钱——他们自己用AI省钱的时候可积极了。
—— 一位娱乐行业律师的评论
Hollywood isn't against AI; they just don't want other people using AI to make money off their IP — they're perfectly eager to use AI to save their own money.
— An entertainment lawyer's comment
Midjourney · Hollywood · AI copyright · Disney · Universal Pictures · Warner Bros · fair use · discovery · AI design tools
Sources · 信源 Sources
本文基于 Dawn Vision 认知引擎处理的 7 个源信号生成,经编辑部人工审核。素材来源:凤凰网科技、IT之家。
This article was generated from 7 source signals processed by the Dawn Vision cognitive engine, with editorial review. Sources: iFeng Tech, IT Home.