Native Foreign Secures AI Copyright for 'Critterz' Short

Native Foreign为AI短片《Critterz》获得版权 电影制作人可通过记录人类参与环节并设计免责声明以保留未来的所有权主张，从而在现行版权法下对AI辅助电影进行登记与保护。 在本次研讨中，Nik Kleverov与律师Lisa Oratz探讨了生成式AI、电影制作与知识产权的交集。Kleverov详细分享了他运用AI工具的历程，从最初的DALL·E 2实验到创作短片《Critterz》，再到将其拓展为长篇故事片。Oratz阐述了当前美国的版权标准，指出人类独创性、富有表现力的提示词以及后期制作剪辑是获得版权保护的关键。他们还提供了切实可行的应对策略，用以降低第三方侵权风险、应对企业赔偿条款，并鼓励独立电影人积极接纳AI创作工艺。 Native Foreign 制作的早期 AI 动画短片《Critterz》（最初基于 DALL·E 2 制作并由 Sora 重制）现已与 Vertigo Films 及《帕丁顿熊》编剧团队合作，正拓展为一部完整的院线动画长片。 (7:09) 在 AI 电影制作中，生成流程主要被视为后期制作任务，促使传统剪辑师转变为跨领域的导演，显著降低了独立创作者实现复杂视觉叙事的门槛。 (2:00) 目前尚无法律判例全盘否定 AI 辅助作品的版权性，创作者应通过修改原始 AI 输出、结合自创手绘或剧本等人类表达，并全程留存创作记录以强化版权主张。 (7:56) 《Critterz》成功获得美国版权局的官方作品登记，其策略在于仅对“法律认定不具版权性的范围”进行针对性免责声明，既规避了审查阻力，又为未来法律演进保留了权利空间。 (16:46)

Generative artificial intelligence offers independent filmmakers unprecedented narrative scale, but securing intellectual property protections requires navigating strict administrative standards of human authorship and proactive risk management. According to filmmaker Nik Kleverov and entertainment attorney Lisa Oratz, demonstrating legal ownership depends on treating AI as an assistive post-production instrument rather than an automated replacement for human creative control. Production Workflows and Scaling Critterz (3:44) Native Foreign began developing the animated short film Critterz in late 2022 using early experiments with OpenAI's DALL-E 2, steering away from dystopian sci-fi tropes to produce an irreverent nature documentary parody. Chief Creative Officer Nik Kleverov approaches AI generation primarily through an editorial lens, arguing that text-to-image and text-to-video tools transform the traditional editor into a director who rapidly assembles narrative concepts without standard physical production constraints. After screening at the Tribeca and Annecy film festivals and being remastered using OpenAI's Sora model, the short is expanding into a feature film in partnership with Vertigo Films. The production has established a London office and hired screenwriters from the Paddington animated series and Paddington in Peru, illustrating how independent studios are integrating conventional screenwriting structures with generative toolchains. Legal Standards for Human Authorship and Control (7:56) Senior Counsel Lisa Oratz clarifies that no binding judicial precedent categorically excludes AI-assisted content from copyright protection. The only primary court ruling to date addressed an applicant who claimed zero human involvement, establishing human authorship as a threshold requirement rather than barring computer-assisted works. Instead, the restrictive standards frequently cited by creators originate from US Copyright Office administrative guidance, which does not carry the weight of judicial case law. The Copyright Office's core criterion is human control over the final output. The agency currently treats text prompts as mere influence, determining that prompting alone lacks sufficient control to confer authorship. However, copyright protection remains available for manual edits to AI outputs, human selection and arrangement of multi-asset compilations (including audio, scripts, and editing choices), and generations guided by expressive inputs such as original hand-drawn sketches. Strategic Disclaimer in Copyright Registration (16:46) Critterz became the first AI-assisted animated short to receive an official US Copyright Office registration certificate by using a tailored application strategy. To satisfy federal disclosure requirements without surrendering legal arguments, counsel disclaimed AI-generated elements only to the extent that those elements are legally uncopyrightable, rather than issuing a blanket disclaimer. This conditional language preserves claims to protection should judicial standards evolve, while introducing ambiguity that discourages potential infringers who might assume AI elements reside entirely in the public domain. To substantiate authorship in such filings, creators must systematically archive their creative choices. Native Foreign logs character illustrations, script revisions, prompt trees, and editorial decisions to produce a verifiable audit trail of human intervention. Mitigating Third-Party Infringement and Commercial Risk (17:37) More than 40 copyright lawsuits have been filed in the United States regarding generative AI, but these actions target model developers over training data scraping rather than downstream users. For filmmakers, the primary legal exposure is output similarity—the risk that an output substantially copies protected existing works. Oratz warns that prompting models to emulate a specific artist's style sharply increases the likelihood of generating infringing material. Filmmakers can mitigate output risk by anchoring generation in original reference sketches, selecting enterprise tools trained exclusively on licensed or public domain data, and utilizing commercial platforms like Adobe Firefly or Freepik that offer enterprise indemnity. Kleverov contends that enterprise developers must assume responsibility for clearing training datasets, provided end-users are not deliberately trying to reproduce protected intellectual property. While the United States strictly conditions copyright on human authorship, other global jurisdictions have recognized computer-generated works under statutory frameworks dating back to the 1980s. Kleverov and Oratz emphasize that AI is not an automated shortcut around craftsmanship, and navigating the evolving international market requires combining traditional filmmaking knowledge with deliberate legal planning.

Native Foreign首席创意官Nik Kleverov与Perkins Coie律师事务所高级法律顾问Lisa Oratz指出，AI生成技术已成为独立创作者与创意机构实现复杂叙事的有力杠杆；创作者若要确立AI生成作品的版权并规避第三方侵权风险，核心在于将传统视听技艺、全流程的人类作者身份记录以及审慎的法律登记策略相结合。 影视创作实践：从概念实验到长片工业化拓展 (3:44) Native Foreign团队自2022年4月起接触OpenAI的DALL·E 2技术，并在同年底启动了动画短片《Critterz》的制作。该片摒弃了早期AI演示中常见的反乌托邦科幻调性，将《地球脉动》（Planet Earth）式的自然纪录片形式与《巨蟒剧团》（Monty Python）式的英式荒诞喜剧融合，通过虚构解说员丹尼斯与红杰克蜘蛛等森林生物的戏谑对话展开叙事。短片在翠贝卡和昂西等国际电影节展映后，又经由OpenAI的Sora模型完成了画面重制。目前，该项目已获得OpenAI部分资助，并与Vertigo Films以及《帕丁顿熊》系列编剧团队签约，在伦敦设立工作室将其改编为院线长片。 Kleverov认为，AI生成本质上是一种让剪辑师承担导演职责的后期制作流程。借助文生图与文生视频工具，创作者能够摆脱传统实体拍摄的制约，将复杂的视听概念迅速落地。但他同时反驳了AI是“一键成片魔法按钮”的论调，强调高质量的商业与叙事作品依然高度依赖传统的镜头语言、剧本创作、配音表演以及大量耗时的精细打磨。 版权归属界定：美国版权局的“控制权”审查标准 (7:56) 针对AI生成内容无法受版权保护的普遍误解，Oratz澄清指出，目前司法层面尚未形成否定AI辅助作品版权的具有约束力的判例。唯一的重大司法判例仅适用于作者声明“零人类干预”的极端情形，确立了人类作者身份是版权保护的基本门槛。现行被广泛讨论的限制性规定多源于美国版权局的行政指导方针，并不具备司法判例的强制力。 美国版权局评估AI辅助作品可版权性的核心指标是创作者对输出成果的“控制权”（Control）。在这一标准下，仅输入文字提示词（Prompts）被视为对系统的外在影响而非直接的法律控制，因而单凭提示词无法获得版权认可；然而，创作者对AI输出进行的后期手工修改、跨媒介素材的筛选编排（汇编作品），或是输入具有版权的人类原创图稿与分镜作为生成参考，均可使作品中凝聚人类独创性的部分获得法定保护。因此，建立详细的创作档案以留存草图、提示词演化版本及剪辑记录，是证明人类作者权的关键举措。 权利主张策略：《Critterz》的有限度免责声明 (14:43) 短片《Critterz》成为首部获得美国版权局正式注册证书的AI驱动动画短片。在递交申请时，制作方明确主张了人类撰写的剧本、配音表演和剪辑编排构成的整体汇编权，但在处理版权局要求放弃非人类生成部分权利的规定时，采取了极具策略性的法律措辞。 与以往部分项目全面放弃所有AI元素的做法不同，Perkins Coie在申请中特别声明“仅在AI生成元素依法不具备可版权性的限度内放弃权利”（only to the extent that they are not copyrightable）。Oratz解释称，这种弹性表述避免了未来版权法判例向有利于AI创作者演进时无法重新主张权利的被动局面，同时也保留了法律上的不确定性，能够有效震慑企图将片中AI元素直接视作公有领域资产的侵权者。该申请最终在未经历反复驳辩的情况下直接获批。 侵权责任区隔：模型训练风险与终端用户输出合规 (17:37) 在探讨AI生成内容侵犯第三方知识产权的风险时，讨论明确区分了“模型训练风险”与“终端输出风险”。目前全美已发起40多起针对AI企业的版权诉讼，但这些案件的被告均是模型开发商而非终端创作者。对于使用商业模型的创作者而言，法律风险主要集中在生成的最终成果是否与第三方享有版权的作品构成“实质性相似”。 为降低侵权概率，创作者应避免输入指定特定艺术家独特风格的提示词（例如“模仿某艺术家风格”），因为此类提示显著增加了输出与训练数据高度相似的概率。创作者应当使用原创图稿引导生成，或选用仅在公共领域及已授权数据集上训练的模型工具。此外，Adobe Firefly与Freepik等平台推行的企业级免责赔偿协议（Enterprise Indemnities），进一步在服务条款中将开发者的抓取责任与用户的恶意提示侵权区隔开来，为合规创作者提供了关键的商业保障。 各国在AI版权领域的法律演进并不同步，例如美国严格坚持人类作者身份，而部分其他国家早在20世纪80年代即设立了计算机生成作品的相关框架。研讨嘉宾总结认为，虽然国际法规与监管环境仍处于动态变化之中，但创作者只要规范落实以人类为核心的创作工作流、审慎筛选合规工具并善用企业免责条款，便完全能够在管理好法律风险的同时拓展AI叙事的边界。

## Background 该视频由 YouTube 频道 Ai4 于 2025 年 8 月 28 日发布。内容呈现了 Native Foreign 首席创意官 Nik Kleverov 与 Perkins Coie 律师事务所高级顾问 Lisa Oratz 的主题对谈，以业内首部采用 OpenAI DALL·E 视觉并正改编为院线长片的开创性 AI 动画短片《Critterz》为案例，深入探讨生成式 AI 时代的影视创意协作、版权保护策略与法律合规风控。 ## TL;DR 对谈围绕 AI 原生短片《Critterz》的创作与长片孵化展开，探讨了从片头剪辑到 AI 导演工作流的演进，展现了独立电影人借助 AI 跨越制作壁垒的可能。针对版权争议，法律专家 Lisa Oratz 深入解析了美国版权局对“人类独创性”的高门槛要求，提出通过表达性提示词（手绘草稿）、深度后期修改、编排与全流程文档留存来最大化版权保护。同时，嘉宾厘清了模型训练与输出端的侵权责任边界，建议创作者规避风格模仿提示词并优先选用具备企业级赔偿（Indemnity）的合规工具，呼吁创作者打破“一键成片”的迷思，积极投身技术驱动的创意文艺复兴。 ## Atmosphere & style 整场对谈氛围轻松幽默而又极具专业深度，创意导演与知识产权律师形成了互补默契的双重视角。Nik Kleverov 以充满激情的实战经历打破了外界对 AI 魔法式生成的神话，Lisa Oratz 则用严谨透彻的法理逻辑为创作者拨开版权迷雾。现场交流节奏明快、干货密集，既有跨界探索的先锋实验感，又充满了对独立电影未来充满信心的务实与乐观。 ## Key takeaways - [02:12](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=132s) Native Foreign 早期从事影视剧主片头设计，其高度提炼和后期属性与生成式 AI 创作高度重合，使剪辑师能够掌握更多导演语汇。 - [04:01](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=241s) AI 动画短片《Critterz》不仅入选翠贝卡与安纳西电影节，还与 Vertigo Films 及《帕丁顿熊》编剧合作孵化为院线长片电影。 - [07:56](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=476s) 美国版权局（USCO）当前采取狭隘的审查标准，认为单纯文本提示词仅构成“影响”而非“控制”，要求具备充分的人类独创性与控制权。 - [10:22](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=622s) 人类原创剧本、后期二次创作以及对 AI 素材的选择与编排，作为不可分割的集合体依然具备极强的可版权性。 - [10:22](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=622s) 使用手绘草图等作为“表达性提示词”能将原创作品的底层表达映射到生成结果中，使该部分受到版权保护。 - [12:19](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=739s) 最大化版权保护的关键在于“最大化人类触点”并建立完整的创作过程文档，详细留存草稿与提示词迭代记录。 - [15:22](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=922s) 《Critterz》采用“仅在不受版权法保护的限度内放弃 AI 生成元素”的声明策略，成功且无争议地通过了美国版权局登记。 - [17:37](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1057s) 现有诉讼均集中在模型训练阶段且针对 AI 开发商，普通终端创作者主要需防范输出端与既有作品构成“实质性相似”的风险。 - [19:53](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1193s) 创作者应避免使用模仿特定艺术家或知名 IP 的诱导性提示词，并优先选用提供企业级侵权赔偿保证（Indemnity）的工具。 - [22:41](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1361s) AI 为独立电影带来了创意复兴，赋能微型团队实现大厂级制作，但高品质创作仍需投入同等精力打磨而非“一键生成”。 - [24:15](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1455s) 全球各司法管辖区对计算机生成作品的法律规定存在差异，通过规范工作流与谨慎选型可有效管理跨境法律风险。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=4s) 开场介绍：Native Foreign 与 Perkins Coie 的跨界合作 对谈在轻松幽默的氛围中开始。Native Foreign 联合创始人兼首席创意官 Nik Kleverov 首先介绍了坐在身旁的法律顾问——来自 Perkins Coie 律师事务所的高级法律顾问 Lisa Oratz。Lisa 调侃道，虽然 Nik 并不总是需要带律师上台，但能够共同探讨这一前沿领域令人兴奋。Lisa 介绍自己的业务深耕于科技、知识产权与娱乐产业的交叉点，服务对象涵盖全球顶尖的科技巨头、前沿初创公司以及独立电影创作者与制片厂。 Nik 随后介绍了 Native Foreign 的发展历程。公司起步于传统的商业广告制作与创意代理混合模式，随后全面拥抱生成式 AI，并在业界创下了多项纪录：包括去年夏天引爆全网的玩具反斗城（Toys "R" Us）首支 AI 品牌广告片，以及与《Vogue》杂志合作打造的首个时尚杂志 AI 案例。此外，他们还为达美航空（Delta）、可口可乐（Coca-Cola）、维珍游轮（Virgin Voyages）等知名品牌打造 AI 营销战役，并致力于开发原创故事内容。 - **对谈嘉宾**： - **Nik Kleverov**：Native Foreign 首席创意官，资深剪辑师、导演与 AI 创作者。 - **Lisa Oratz**：Perkins Coie 高级法律顾问，专注知识产权、科技与娱乐法。 - **代表项目与合作方**：Toys "R" Us 首支 AI 商业片、Vogue 杂志 AI 特辑、Delta、Coca-Cola、Virgin Voyages 等。 --- ### [02:12](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=132s) 创意历程：从片头设计与剪辑到 AI 导演模式 谈及自己的创作背景，Nik 分享了成长故事：小时候他就像电影《美国丽人》（American Beauty）里那个痴迷拍摄空中漂浮塑料袋的孩子一样，整天拿着摄像机拍摄各种画面。他的父亲曾是苏联的知名艺术家，因创作批评苏联体制的画作而入狱，最终流亡出逃；但 Nik 笑称自己完全没有遗传到绘画天赋，连画一匹火柴人小马都做不到，因此只能通过镜头和剪辑去讲述故事。 Nik 指出，Native Foreign 早期从事了大量影视剧主片头（Main Title Sequence）的设计与制作，这种工作要求创作者在 30 秒到 1 分钟之内高度提炼全剧的核心主题与视觉隐喻。他发现，使用 AI 进行内容创作与片头制作有着惊人的相似性——它在本质上是一项高度依赖后期的工作，“剪辑师现在成为了导演”（the editor has now become the director）。作为一名剪辑出身的创作者，Nik 认为绝大部分叙事魔力最终都是在剪辑台上合成的，而 AI 工具极大地解放了个人的视觉叙事潜能。 ```mermaid flowchart LR accTitle: 从传统后期剪辑到AI导演工作流 subgraph S1["传统影视与片头制作"] A["高维概念提炼"] --> B["后期剪辑/视觉重构"] end subgraph S2["生成式 AI 创作流"] C["AI 视觉与素材生成"] --> D["剪辑师主导导演语汇"] end B -.->|"技能与工作流迁移"| D ``` --- ### [04:01](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=241s) 原创 AI 短片《Critterz》的诞生与长片电影孵化 Nik 详细回顾了开创性 AI 动画短片《Critterz》的制作始末。早在 2022 年 4 月（在 Midjourney 和 ChatGPT 尚未席卷行业之前），团队就开始试用 OpenAI 的 DALL·E 2 进行图文探索，并于 2022 年底正式投入制作。当时多数科技演示片都在展示反乌托邦或科幻题材，而他们反其道而行之，向 OpenAI 提案了一个荒诞搞笑的森林生物世界，其风格被 Nik 形象地描述为“大卫·爱登堡（David Attenborough）的《地球脉动》一头撞上了蒙提·派森（Monty Python）的喜剧小品”。 短片虚构了大卫·爱登堡的邻居“丹尼斯”（Dennis）作为全知旁白，试图严肃记录未被人类踏足的神秘森林，却遭遇了各种性格古怪的生物：如每天精准睡 23.6 小时的放哨小兽、以及在树上直接与旁白拌嘴并吐槽拍摄侵犯隐私的红色蜘蛛“弗兰克”（Frank）和“布鲁”（Blue）。 《Critterz》在完成制作后入选了翠贝卡电影节（Tribeca）与安纳西国际动画电影节（Annecy）等顶级展映，近期更使用 OpenAI Sora 进行了全新重制。更令人振奋的是，这一最初由 OpenAI 资助的独立 AI 短片正在被改编为院线长片电影，并由《帕丁顿熊》（Paddington）动画系列编剧执笔剧本，实现了从 AI 原生实验短片到主流好莱坞 IP 孵化的跨越。 --- ### [07:56](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=476s) AI 生成内容的版权困境与美国版权局（USCO）立场 针对创作者普遍关心的“AI 生成作品能否获得版权保护”问题，Lisa Oratz 从法律实务角度进行了深度解析。她澄清，目前大众常说的“AI 作品不受保护”在法律层面上并不准确，因为现阶段并无充足的司法判例来确立明确规则。迄今为止唯一的法院判决仅确立了“必须包含人类独创性（Human Authorship）”，但由于该案的原告主张作品完全由机器自主生成且无人类参与，因此法院并未界定到底需要“多少”或“何种程度”的人类干预才算达标。 Lisa 强调，目前流传的严苛观点主要源于美国版权局（Copyright Office）的行政审查指南，而版权局的意见并不能直接等同于法律，尽管其在作品登记和司法参考上具有很强的影响力。目前美国版权局对 AI 作品的受保护性采取了极其狭隘的认定标准： - **控制权标准（Control vs. Influence）**：版权局设立了极高的人类独创性门槛，核心在于人类创作者是否能够对最终输出结果实现实质性的“控制”（Control）。即便创作者撰写了极其详尽、复杂的提示词（Text Prompts），版权局仍认为提示词仅构成“影响”（Influence）而非最终控制，因此不予认可。 - **可受保护的人类贡献路径**： 1. **后续修改与二次创作**：在 AI 生成的素材基础上，由人类进行的实质性修改、绘图或后期加工部分享有版权。 2. **选择与编排（Selection and Arrangement）**：借鉴传统版权法原则，如果创作者对不受保护的单个 AI 元素进行了独特的挑选、组织与剪辑编排，该整体汇编作品同样具备获得版权保护的可能。 ### [10:22](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=622s) 表达性提示词与 AI 影视作品的可版权性 在探讨生成式 AI 产出的版权问题时，Lisa Oratz 指出，几乎没有人只凭一条单一提示词就能制作出一整部电影。创作者通常会将多种元素组合在一起，并与剧本等由人类创作的元素深度结合。即便美国版权局（US Copyright Office）持有较为狭窄的审查立场，他们也明确承认这种人类元素的编排与组合是具备可版权性的（protectable）。当前的法律核心问题在于如何界定哪些细分组成部分是不受保护的。 除了常规的文本提示词（text prompts），版权局还认可一种所谓的“表达性提示词”（expressive prompt）。如果创作者不单单输入文字，而是以手绘草图、原始图像等具备版权的作品作为输入起点，版权局承认只要这些原始创作在最终生成结果中得以体现，该部分就会受到版权保护；尽管 AI 工具新增的衍生细节可能无法直接确权，但底层的原始创作表达依然受法律保护。 在电影制作实践中，由于涉及大量人类编剧、后期修改以及对 DALL·E 等工具生成图像的复杂拼接组合，外界很难将纯粹的 AI 生成碎片生硬地剥离出来并随意商用。因此在实操层面上，电影创作者无需对此过度担忧，只要整体作品凝聚了足够的人类构思与编排，整部电影作为不可分割的创作集合体就具备极强的版权抗辩力。 ```mermaid flowchart TD accTitle: AI创作元素与版权保护路径 subgraph HumanInput["人类输入与前置创作"] A["人类原创剧本 / 分镜"] B["表达性提示词<br/>(手绘草图 / 原始图像)"] C["详细文本提示词"] end subgraph AIProcess["AI 辅助生成与加工"] D["DALL·E 等 AI 工具生成"] E["人工后期修改 / 画面拼接组合"] end subgraph CopyrightOutcome["版权认定结果"] F["可获版权保护整体作品<br/>(含表达性提示词映射部分)"] G["未受保护的纯 AI 原生细碎元素"] end A --> E B --> D C --> D D --> E E --> F D -.-> G ``` --- ### [12:19](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=739s) 最大化版权保护的策略与长篇电影开发 针对创作者如何在使用 AI 过程中最大化作品的可保护性并顺利获取版权登记，Lisa Oratz 建议核心策略是“最大化人类触点”（maximize human touch points）。创作者切忌直接拿 AI 工具刚跑出来的原始素材不加改动地发布，而是应当主动对画面进行重绘、修改和组合，并详尽记录人类决策是如何塑造和控制最终视觉输出的；同时，尽量使用辅助性工具将拥有版权的原创草图作为输入源。 Nik Kleverov 随后现场互动询问台下有多少影视同行，并分享了短片《Critterz》的重大进展：Native Foreign 正在与知名制片公司 Vertigo Films 合作，将其改编并制作成一部完整的长篇院线电影。尽管 Native Foreign 以往深耕品牌故事叙述领域，但目前已在英国伦敦设立办公室组建长篇电影团队。团队在长篇制作中严格贯彻“人类作者身份优先”（human authorship first）原则，建立完备的归档体系，全程留存提示词记录、角色原始设计及喂给 AI 的插画手稿，为后续长篇作品的版权登记筑牢证据链。 - **最大化人类触点**：避免直接采用 AI 原生输出，必须经过人类后期修改、多层合成与剪辑。 - **过程文档化（Documentation）**：详尽留存人类决策过程、手绘草稿和提示词迭代记录，以备向版权局清晰说明人类独创性贡献。 - **《Critterz》院线长片计划**：由 Native Foreign 联合 Vertigo Films 共同打造，设立伦敦制作团队推进长篇动画制作。 --- ### [15:22](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=922s) 《Critterz》巧妙的版权登记方案与免责声明策略 《Critterz》成功获得了美国版权局的版权登记，成为业内首批以 AI 视觉为主导并顺利获批版权的影片案例。Lisa Oratz 透露，她此前从不担心电影整体无法注册，因为其中包含人类撰写的剧本及大量后期修改。当时真正的法律博弈在于：如何既遵守版权局要求披露 AI 元素的规定，又不主动放弃为 AI 生成元素争取版权的潜在权利。 按照版权局的标准流程，申请人必须标明作品中哪些部分由 AI 生成、说明人类的独创贡献，并主动声明放弃（disclaim）非人类创作部分的权利。许多其他 AI 相关的注册申请直接声明放弃所有 AI 元素，但 Native Foreign 与法律团队采取了一种极具技巧性的表达方式，在提交时写明：“仅在不受版权法保护的限度内，放弃 AI 生成元素”（disclaimed the AI generated elements to the extent not copyrightable）。这种“留白”策略既未直接与版权局产生僵局冲突，又锁定了法律未来演变的解释空间，甚至让潜在侵权者心存顾忌；最终，美国版权局未经任何反复质疑与质询，直接核准了该申请。 --- ### [17:37](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1057s) 第三方侵权风险剖析：训练集争议与用户输出端防范 针对创作者普遍担忧使用 AI 会不会无意中侵犯第三方知识产权的问题，Lisa Oratz 分析了当前的诉讼格局。截至当时，全美已有超过 40 起与生成式 AI 相关的版权侵权诉讼，但所有案件的起诉对象全部是 AI 模型与工具开发公司，目前尚无针对普通 AI 终端使用者的诉讼。 在法律定性上，侵权风险主要被划分为两大范畴： 1. **模型训练阶段（Training Process）**：涉及海量抓取和复制他人版权数据用于训练模型。这属于模型研发公司或自行训练私有大模型的大型机构面临的风险，与绝大多数直接使用现有工具的影视创作者无关。 2. **生成输出阶段（Output）**：属于终端使用者真正需要防范的核心风险，即 AI 最终生成的画面或素材是否与训练集中的既有版权作品构成“实质性相似”（substantially similar）。 Lisa 提醒创作者，输出端侵权风险最高发的场景往往是使用了类似“以某某艺术家风格创作 X”（in the style of...）这类刻意模仿特定创作者风格的提示词。尽管在美国版权法体系下单纯的“艺术风格”（style）本身不受版权保护，但此类提示词极易诱导模型生成在具体表达上与原作高度雷同的内容，从而显著放大侵权法律风险。 ### [19:53](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1193s) 规避侵权风险的实用技巧与工具公司的法律赔偿保证 在讨论如何降低生成式 AI 内容的侵权风险时，Lisa Oratz 强调了几个关键要点。首先是回归到“表达性提示词”（Expressive Prompts）与草图输入的重要性——正如 Nik Kleverov 所介绍的创作流程，若从创作者本人的手绘草图出发，并将其作为引导模型的核心输入，那么生成出的最终画面将极大程度受到创作者草图的直接影响，从而大幅降低无意中复现他人受版权保护内容的概率。此外，在撰写提示词时，务必避免使用任何旨在诱导模型生成特定已有作品或角色相似外观的词汇。 其次，选择哪家 AI 工具开发商至关重要。不同厂商在模型训练数据与过滤机制上采取了截然不同的策略：有的公司完全仅使用公有领域（Public Domain）或已获得合法授权的素材进行训练；尽管在模型训练端对于“合理使用”（Fair Use）存在非常充分的法律辩护理由，但对于风险承受力较低的机构，选择这类来源透明的工具能显著压低侵权顾虑。更重要的是，目前各大厂商普遍在企业级协议（Enterprise Agreement）中提供侵权赔偿条款（Indemnity）。创作者在采购和使用工具时，应当仔细审查授权许可中的赔偿范围与免责限制。 ```mermaid flowchart TD accTitle: AI创作版权风险控制流程 subgraph 输入与控制 ["创作输入端 (创作者控制)"] A["原创草图 / 概念设计"] --> B["表达性提示词 (避免指名特定IP)"] end subgraph 工具与模型 ["工具筛选端 (厂商能力)"] B --> C{"选择合规 AI 工具"} C --> D["授权数据/公用领域训练模型"] C --> E["企业级侵权赔偿保证 (Indemnity)"] end subgraph 产出与风控 ["最终成果 (法律与版权)"] D --> F["降低实质相似与侵权风险"] E --> F F --> G["可主张人类著作权的最终影片资产"] end ``` 从创作者的角度出发，Nik 坦言自己一直认为清权责任本就应当由工具提供方承担——只要创作者没有恶意尝试去生成米老鼠（Mickey Mouse）并谎称是自己原创的，而是使用了安全的提示词、输入了自己绘制的图像，那么工具方理应提供保障。他非常欣慰地看到，自 Adobe 凭借 Firefly 率先推出赔偿保证后，像 Freepik 等越来越多的工具厂商都在企业级合作中跟进了侵权赔偿机制。Lisa 补充道，这种区分是合理的，因为在法律上需要明确区分两类问题：一类是由于模型训练方式本身带来的问题，另一类则是由于用户恶意提示和生成导致的侵权纠纷。 - **原创底稿引导**：使用创作者手绘草图作为核心输入，能最大程度稀释训练集数据对输出结果的偶然雷同风险。 - **提示词合规**：杜绝诱导性提示词，不主动生成知名商业 IP。 - **企业级赔偿（Indemnity）**：企业端用户应优先争取具有法律免责与赔偿承诺的商用授权工具（如 Adobe Firefly、Freepik 等企业方案）。 --- ### [22:41](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1361s) 独立电影的文艺复兴与 AI 时代的创作实践 当被问及对想要涉足 AI 但心存顾虑或不知从何入手的电影人有何建议时，Nik Kleverov 分享了他与众多年轻学生及资深电影制作人的交流经历。他提到，目前很多高校学生陷入了悲观情绪，甚至觉得自己“没必要再上学了，因为一切都被 AI 颠覆了”。然而，Nik 对此持极其乐观的态度，他认为影视行业正迎来一场“不可思议的创意文艺复兴”，尤其是对于独立电影（Independent Cinema）制作而言。 在 Nik 看来，小型团队现在完全可以凭借天马行空的想象力，将传统影视制作的专业知识、审美功底与前沿 AI 工具的生成能力深度融合，最终独立完成过去只有好莱坞大厂才能制作的作品。他甚至开玩笑感叹道：“真希望我现在只有 13 岁，能伴随着这些工具一起长大。”对于当代创作者，他给出的建议是：充分利用手头拥有的丰富工具，动手去做任何能做的作品。 与此同时，Nik 澄清了外界对 AI 的过度神话。媒体上铺天盖地的抓人眼球的标题让很多人以为 AI 是一键搞定的魔法，但实际动手后就会发现事情绝非如此简单。在 Native Foreign 为客户交付的高品质商业与影视项目中，团队耗费的精力和时间其实与传统制作方式相差无几——AI 确实带来了效率提升与成本优化，但它绝对不是“按下一个魔法按钮，然后‘唰’的一下成品就出来了（not a magic button that you press and voila the thing happens）”。创作者需要做的是全身心投入实践，跟上每天层出不穷的技术演进，并竭尽全力讲好自己想讲的故事。 - **独立制作赋能**：微型团队借助 AI 弥补预算与人员短板，使高概念独立电影的规模化制作成为可能。 - **传统技艺与 AI 结合**：AI 并非替代电影专业知识，而是将传统导演、美术、剪辑的掌控力放大到生成流程中。 - **摒弃“一键生成”幻想**：高品质项目依然需要严谨的工时投入和精细打磨。 --- ### [24:15](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1455s) 全球版权法差异与理性管理法律风险 Lisa Oratz 在最后总结中提醒创作者，讨论中提及的很多版权规则主要基于美国现行的法律框架（如强调必须具备人类作者身份 Human Authorship）。然而在跨国协作或全球发行时必须意识到，各国对 AI 生成内容的法律规定存在显著差异——例如某些国家自 20 世纪 80 年代起就在法律中确立了关于“计算机生成作品”（Computer-Generated Work）的保护概念，并不完全与美国当前的判例相同。 面对当前充满不确定性的法律环境，Lisa 认为电影人不必因噎废食。版权保护性（Protectability）和侵权风险（Infringement）固然是前期规划和制作流程中必须严肃对待的问题，但在电影制作领域，这些风险完全不是不可逾越的障碍。通过审慎选择合规的工具平台、构建清晰的人工介入与草图工作流，绝大多数法律风险都可以得到有效控制与管理。 Lisa 鼓励所有创作者勇敢迈出第一步，“不要害怕去涉足、去尝试（not to be afraid of dabbling in and dipping your toe into the pool）”，关键在于带着清晰的风控意识去规划创作流程，以最低的风险拥抱新技术。最后，Nik 与 Lisa 在现场倒计时提醒中向全场观众致谢，并邀请大家在会后继续展开交流。 - **跨国版权差异**：不同司法管辖区对人类作者身份的要求不同，全球发行业务需综合考量多国法律。 - **风险可控性**：版权不确定性不应成为阻碍创作的理由，规范工作流与技术选型足以化解大部分合规挑战。 ## Notable quotes > [02:12](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=132s) the editor has now become the director — Nik Kleverov > [04:01](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=241s) 大卫·爱登堡（David Attenborough）的《地球脉动》一头撞上了蒙提·派森（Monty Python）的喜剧小品 — Nik Kleverov > [07:56](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=476s) 必须包含人类独创性（Human Authorship） — Lisa Oratz > [12:19](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=739s) 最大化人类触点（maximize human touch points） — Lisa Oratz > [12:19](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=739s) 人类作者身份优先（human authorship first） — Nik Kleverov > [15:22](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=922s) 仅在不受版权法保护的限度内，放弃 AI 生成元素（disclaimed the AI generated elements to the extent not copyrightable） — Lisa Oratz > [17:37](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1057s) in the style of... — Lisa Oratz > [22:41](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1361s) 不可思议的创意文艺复兴 — Nik Kleverov > [22:41](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1361s) 真希望我现在只有 13 岁，能伴随着这些工具一起长大。 — Nik Kleverov > [22:41](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1361s) not a magic button that you press and voila the thing happens — Nik Kleverov > [24:15](https://www.youtube.com/watch?v=rU1rvE9-zH8&t=1455s) not to be afraid of dabbling in and dipping your toe into the pool — Lisa Oratz ## Mentioned Nik Kleverov, Lisa Oratz, Chad Nelson, Native Foreign, Perkins Coie, OpenAI, DALL·E 2, Sora, Critterz, Toys "R" Us, Vogue, Delta, Coca-Cola, Virgin Voyages, Vertigo Films, Midjourney, ChatGPT, Adobe Firefly, Freepik, 翠贝卡电影节 (Tribeca), 安纳西国际动画电影节 (Annecy), 帕丁顿熊 (Paddington), 美国版权局 (US Copyright Office), 大卫·爱登堡 (David Attenborough), 蒙提·派森 (Monty Python), 美国丽人 (American Beauty), 米老鼠 (Mickey Mouse) ## Video description Nik Kleverov, Chief Creative Officer, Native Foreign Lisa Oratz, Senior Counsel, Perkins Coie Abstracts Join Native Foreign Chief Creative Officer Nik Kleverov, OpenAI Creative Specialist Chad Nelson, and Perkins Coie IP, technology, and entertainment attorney Lisa Oratz in conversation as they explore the transformative potential of AI in fostering creativity and offer insights on how filmmakers can navigate an uncertain legal landscape to protect their creative works. Using the award-winning animated short "Critterz" as a case study – which marked a groundbreaking achievement as the first AI film to seamlessly integrate visuals generated by OpenAI’s DALL•E and is being turned into a full-length feature film – the discussion will delve into the human side of generative AI, highlighting the collaborative process and the unique challenges and opportunities it presents. They will also explore copyright issues surrounding the use of generative AI. This session promises to provide a comprehensive and thought-provoking exploration of the intersection between AI, creativity, and intellectual property for a broad audience. [Source](https://www.youtube.com/watch?v=rU1rvE9-zH8)

[2:12] Nik Kleverov: the editor has now become the director [4:01] Nik Kleverov: 大卫·爱登堡（David Attenborough）的《地球脉动》一头撞上了蒙提·派森（Monty Python）的喜剧小品 [7:56] Lisa Oratz: 必须包含人类独创性（Human Authorship） [12:19] Lisa Oratz: 最大化人类触点（maximize human touch points） [12:19] Nik Kleverov: 人类作者身份优先（human authorship first） [15:22] Lisa Oratz: 仅在不受版权法保护的限度内，放弃 AI 生成元素（disclaimed the AI generated elements to the extent not copyrightable） [17:37] Lisa Oratz: in the style of... [22:41] Nik Kleverov: 不可思议的创意文艺复兴 [22:41] Nik Kleverov: 真希望我现在只有 13 岁，能伴随着这些工具一起长大。 [22:41] Nik Kleverov: not a magic button that you press and voila the thing happens [24:15] Lisa Oratz: not to be afraid of dabbling in and dipping your toe into the pool

[2:00] Nik Kleverov: It's a post-production job primarily where the editor has now become the director. And as someone that's come up as an editor... everything kind of gets put together in the edit and now we're really seeing that come to life. [3:44] Nik Kleverov: a David Attenborough Planet Earth documentary crashing into a Monty Python sketch. [5:18] Dennis: Oh dear. I've eaten the wrong berries. [5:18] Blue: Nope. It ain't the berries... I'm Frank, and I'm Blue, which is weird cuz I'm a red spider.

[0:04] What's shaking everybody? How's it [0:05] going? [0:07] It's 3:50 p.m. So, we're right on track. [0:11] We're all feeling great in here. So, [0:14] thank you for being here to talk with us [0:17] about Critters and copyright. Critters [0:20] is a really fun AI film we made. It's [0:22] actually one of the first AI animated [0:24] shorts that was ever made. And we've got [0:26] a pretty fun story. I'm sitting here [0:28] next to my lawyer uh Lisa from Perkins [0:32] Kuey and uh I will let her introduce [0:35] herself. [0:36] >> Hi. And it's not that he needs his [0:38] lawyer to do this, but um it's really [0:40] fun. I've had the fortunate uh pleasure [0:42] of working with Nick. And my practice is [0:44] a combination of technology, [0:46] intellectual property, and [0:48] entertainment. And I do work in all [0:50] those areas, but where it intersects [0:51] such as AI and entertainment and [0:53] copyright is really my sweet spot. and I [0:56] represent companies all the way up to [0:57] the biggest technology companies in the [0:59] world to cutting edge startups and [1:02] creative uh filmmakers and studios. So [1:05] I'm very fortunate that I get to do this [1:07] and to work with folks like Nick. [1:09] >> Thank you. And I'm Nick Cleveroff. I'm [1:11] chief creative of a company called [1:14] Native Foreign. We started as a bit of a [1:16] traditional commercial creation company [1:20] uh agency production company hybrid and [1:22] we've really delved uh really into AI [1:26] and we've we're behind several firsts in [1:28] the AI world. We did the first AI brand [1:31] film commercial. It went very viral last [1:33] summer for Toys R Us. We did uh the [1:35] first instance of AI in a fashion [1:37] magazine with Vogue. We did a lot of a [1:40] lot of first and we're continuing to do [1:42] AI things with both brand partners as [1:45] well as original stories which is what [1:47] I'm excited about because you know we [1:49] always we all want to tell original [1:51] stories and now thanks to the AI tools [1:54] creators like myself and our company we [1:56] we have an opportunity to do so. [1:59] >> Yeah. So Nick, as one of those really [2:00] early pioneers in the area of AI and [2:03] film making, I'm hoping you can tell us [2:04] a little bit about your creative journey [2:06] with AI and how it's really kind of [2:08] helped you realize your creative [2:10] potential. [2:12] >> Yeah, so I have always been a creative [2:16] like as a kid I was that American beauty [2:18] kid that would just film everything with [2:20] the video camera, you know, like a [2:21] plastic bag floating in the sky. Uh my [2:24] father was a famous artist who actually, [2:26] you know, was in the Soviet Union. He [2:29] painted things that was critical of the [2:30] Soviet government to which you know he [2:32] had to escape the Soviet Union. It was a [2:34] whole thing. He was in jail all that [2:35] stuff. Um however I cannot paint like at [2:38] all. Like if you asked me to paint a [2:40] stick figure pony I couldn't do it. So [2:42] I've been telling stories in my own way [2:43] for a long time and because of the AI [2:46] tools now I feel like I'm able to [2:48] explore that fuller. You know, we have a [2:52] a boutique agency that creates a lot of, [2:55] you know, we came from main title [2:56] sequences, which is, you know, you're [2:58] distilling the themes or ideas of a se a [3:00] show down into a minute or 30 seconds or [3:03] less. And when you're creating with AI, [3:05] I've actually noticed it's really [3:07] similar to doing a title sequence job. [3:10] It's a post-p production job primarily [3:12] where the editor has now become the [3:14] director. And as someone that's come up [3:16] as an editor for a majority of my early [3:19] career, you know, I'm not mad about that [3:21] because I think everything kind of gets [3:22] put together in the edit and now we're [3:24] really really seeing that come to life. [3:26] So the ability to now use AI to explore [3:30] both brand campaigns um for Delta, for [3:33] CocaCola, for Virgin Voyages, as well as [3:37] doing our own original shorts, Critters, [3:39] which is what we're here to talk about [3:41] today. Uh it's a it's an exciting time. [3:44] >> Yeah. And for those of you who haven't [3:45] seen Critters, um it's a really was [3:47] groundbreaking project. One of the [3:49] first, as Nick said, to use uh AI [3:52] dominant uh visuals. And so we're going [3:55] to show a short clip of critters and I'm [3:58] hoping you can tell us a little bit [3:59] about the process of creating critters. [4:01] >> Yeah. Can we roll the tape back there? [4:03] We got a short little twominut or less [4:05] uh piece. Okay. While we wait for that, [4:08] I will tell you about Critters and then [4:10] uh it'll it'll come on. So, Critters was [4:13] one of the first AI animated shorts. We [4:16] started working on it in late 22. We [4:19] started playing around with Dolly 2 in [4:22] April of 2022. So, this was pre-made [4:25] journey, pre-hat, GPT, all that stuff. [4:27] When we realized we can use image to, [4:30] you know, text to image to create a [4:31] bunch of stuff, that was a pretty [4:33] exciting time. So, we created this [4:34] entire world of woodland critters. And [4:37] we actually pitched Open AI. Hey, you [4:40] know, there's all these like tech demos [4:41] going on about like dystopian sci-fi [4:43] futures, but what about like this cute [4:45] little, you know, irreverent comedy, [4:48] which is essentially a David [4:50] Atenboroough planet Earth crashing into [4:52] a MontiPython sketch. So, this omnisient [4:54] narrator walks us through this world of [4:56] critters, but he's kind of all wrong. At [4:59] one point, you know, when we did it, uh, [5:01] it did a lot of festivals. Went to [5:03] Tribeca, it went to Anacey, it did a [5:04] huge festival run, and then recently we [5:07] remastered it with, um, with Sora. So, [5:09] I'm going to just going to play the [5:10] little clip there. [5:18] >> Here we are, Planet Earth. It's [5:21] impossible not to appreciate the sheer [5:23] grandeur of our world. And yet there [5:26] remains one forest unexplored by humans. [5:30] A forest filled with life, with [5:33] creatures living in the burrows of the [5:35] ferns, the branches of the trees, by the [5:39] flowing streams and the mossy bogs. I'm [5:43] David Atenburgh's neighbor, Dennis, and [5:46] welcome to a forest filled with little [5:50] critters. Our first encounter is this [5:55] curious fellow with the eyes of a hawk [5:58] and the ears of a bat. He stands calmed [6:01] for the many creatures of these woods. [6:04] Now, this astonishing creature is [6:07] thankful for the lookout for it sleeps [6:09] 23.6 hours a day. Specific, I know, but [6:13] witnessing their subconscious adventures [6:16] is truly adorable. [6:20] Looking upward, we find the red jack [6:23] spider. It's hairy, distinctive for it. [6:26] Excuse me. [6:27] >> Wait, I'm sorry. Who is speaking? [6:29] >> I'm speaking to you. [6:31] >> Oh dear. I've eaten the wrong berries. [6:33] >> Nope. It ain't the berries. [6:35] >> What? [6:35] >> Who are you? [6:36] >> I'm Frank [6:37] >> and I'm Blue, which is weird cuz I'm a [6:40] red spider. [6:41] >> I could have call you Frank or Blue. [6:43] This is a science documentary. [6:45] >> Science? You think secretly filming [6:48] someone while they sleep is science? [6:51] >> What? Yes. [6:52] >> Well, I know it's creepy. [6:55] >> So creepy. [6:56] >> Listen, I can assure you I'm here as an [6:58] award-winning documentarian behest by [7:01] his majesty. [7:02] >> Oh, so you're an English colonialist? [7:04] >> No, no, we're not here to colonize, [7:06] >> right? Pretty soon you'll have us [7:08] talking all day. [7:09] >> Okay, so yeah, so kind of keeps going on [7:12] and on like that. So I think the the [7:14] interesting thing about seeing it in [7:16] Sora is you get to see you know just in [7:19] 18 months how much the technology had [7:22] advanced. And the kind of exciting thing [7:25] is this little random short film that we [7:27] made that open did you know um pay for [7:30] that then went on to do this festival [7:32] run is now becoming a feature film. And [7:35] not only that it's it's written by the [7:36] writers Paddington uh Paddington Bear. [7:39] So, the Paddington and Peru, the [7:40] animated series, and you know, we've [7:43] taken this little AI native short and [7:45] it's now becoming a much bigger thing um [7:49] because of the tools, because of the [7:50] human creativity behind, and you know, [7:53] we're here to talk a little bit about [7:54] that. [7:56] >> Yeah. And you know, I think one of the [7:59] questions that we get a lot uh in this [8:01] area is, you know, how do I protect this [8:03] work? because there's a lot of concern [8:05] about copyright protection if we use AI [8:07] generated tools. Um, but one thing to [8:10] realize is we don't really know what the [8:12] law is because when I hear people say, [8:14] oh, you know, AI generated work isn't [8:15] protectable. Um, it's important to [8:18] realize there's really the legal [8:19] landscape is unclear because there's [8:21] really only one case on this. And that [8:23] case just said that there has to be some [8:25] human authorship. What we don't know is [8:28] how much or what type of human [8:30] authorship because that um that [8:33] particular artist claimed no human [8:35] authorship at all. They were trying to [8:37] argue that against that requirement. So [8:40] really until we have some actual case [8:42] law, we really don't know what is [8:44] required. But when you hear people talk [8:46] about, you know, AI, you know, if it's [8:47] generated stuff isn't protected, what [8:50] you're really hearing is reflecting the [8:51] views of the copyright office, and even [8:54] that's a bit of an overstatement. Um but [8:56] it's important to note that first of all [8:58] what the copyright office says is not [9:00] law. Now it's influential because they [9:02] do decide at least initially what gets [9:04] registered and they're very you know can [9:07] be influential on courts but until we [9:09] have cases again it's you know we really [9:11] can't say uh but what does the copyright [9:14] office say because that's you know what [9:15] we have right now and I will say that [9:17] they have taken a pretty narrow view of [9:20] protectability. [9:21] um in particular [9:24] um they set a very high bar for the [9:26] level of human uh authorship required [9:29] and what they say they're looking for is [9:31] control. So they want you to show that [9:33] the output was really controlled by a [9:36] human in order for it to be protectable. [9:39] Um and so for example things like text [9:41] prompts even though there's an argument [9:43] that a human creates a prompt and that [9:44] influences the output you know that that [9:47] should be sufficient human authorship [9:49] but the copyright office takes the [9:50] position that even with very very [9:52] detailed prompts that's not sufficient [9:55] uh because again it influences but it [9:57] doesn't control the outcome. Um now that [10:00] doesn't mean that they don't recognize [10:02] anything in AI generated works. So for [10:05] example they have made it very clear [10:07] that if you take an AI generated output [10:10] and you modify it those modifications [10:12] are protectable. Uh there's also another [10:14] kind of common principle in copyright [10:16] law they recognize which is that if you [10:18] take unprotected elements and you select [10:21] and arrange them that can be [10:22] protectable. So you know most people I [10:25] know don't take a you know like a prompt [10:27] and create a film based on one prompt. [10:29] They're combining elements uh with other [10:32] elements. they're combining it with [10:34] human created elements like scripts. Uh [10:37] so you know if you do that the copyright [10:39] office even in their narrow view has [10:40] said that combination is protectable. Uh [10:43] the issue really is figuring out what [10:46] pieces aren't protectable. And you know [10:49] a couple other things I've said that I [10:51] think are kind of interesting to keep in [10:52] mind. Uh and that is aside from text [10:55] prompts there's something they call an [10:56] expressive prompt. So if you start not [10:59] just with a text prompt but you start [11:01] with say an image a sketch uh that is a [11:04] protectable work and they recognize that [11:06] to the extent that that is reflected in [11:08] the output that will be protected. Now [11:10] the new stuff that the AI tool may add [11:13] may not be uh but the underlying um tool [11:18] excuse me underlying uh generated [11:21] content from excuse me the original [11:23] content is protectable. Um so I think [11:27] you know the there's a lot of questions [11:29] around that but if you think about how [11:31] this impacts film in particular you know [11:34] it's not just generating an image as I [11:37] said nobody just creates a whole film so [11:39] they're very good arguments that the [11:41] film as a whole are protectable and the [11:42] issues are around those individual [11:44] elements so if somebody could figure out [11:46] you know what originally came out like [11:48] in this film could you tell what you [11:51] know what came out as you can see [11:52] there's a lot of human uh there was a [11:54] human script they uh these images from [11:57] Dolly were modified and combined. It [11:59] would be very hard to pull out those [12:02] human generated excuse me the AI [12:04] generated pieces from the film and be [12:06] able to say we can use this. So as a [12:08] practical matter um you know it's not [12:11] something that I think filmmakers need [12:13] to overly worry about and we'll talk a [12:16] little bit later about you know [12:17] strategies you can use. One of the [12:19] things we've talked about a lot is you [12:22] know what creators can do in the process [12:24] of using AI to maximize the [12:26] protectability and the ability to obtain [12:29] the copyright registration right so for [12:31] films made using you know in whole AI or [12:34] or partly AI you know what is the advice [12:37] that you may have [12:39] >> yeah so things that we talk about with [12:41] companies like yours and other clients [12:42] are to really maximize those human touch [12:45] points you know you don't want to take [12:47] something right as it comes out of a [12:49] tool and just put it out there. I mean, [12:50] you can, but it would be difficult to [12:52] protect. So, you want to make sure that [12:54] you're modifying images, you're [12:55] combining content, and you're really [12:58] doing everything you can to document how [13:00] what a human being decides is shaping [13:03] and controlling that output. Um, another [13:06] thing that you can do is use more of [13:08] those assistive um, uh, type of tools so [13:12] that if you have a copyrighted content [13:15] and you're using that as input like a [13:17] sketch, you're going to have, you know, [13:19] much more clear rights to that output. [13:22] So, you know, again, there's a lot of [13:24] things you can do to really maximize um, [13:26] that protection. [13:28] >> Are there a lot of filmmakers in this [13:30] room? I'm just curious like who is a [13:32] filmmaker in this room? Okay, there's a [13:34] few of us holding it down in here. [13:36] >> Very nice. All right. Interesting. [13:39] >> Yeah. Now, you know, some of the other [13:40] exciting news that you may or may not [13:42] have heard about Critters is that Native [13:44] Forum is working with Vertigo Films to [13:46] turn it into a full-length feature film. [13:49] Uh, so I wanted to have you talk a [13:51] little bit about um about that and and [13:54] what's to come. [13:55] >> Yeah. So again, it's it's this kind of [13:57] short that we made that then got [13:59] obviously attention and then we [14:01] partnered up with someone that's made a [14:03] lot of movies. You know, we come from a [14:05] brand storytelling world, so long form [14:08] content has not been our foray yet. But [14:10] I will say we've got a office popped up [14:12] in London and I just was over there, you [14:15] know, checking on the team and, you [14:16] know, it's really interesting to see how [14:19] now the team is putting together the [14:21] film and we're we're doing exactly what [14:24] you just said, which is that kind of [14:25] human authorship first, keeping really [14:27] good records of our prompts, our [14:30] character design, you know, the [14:32] illustrations that then we're feeding [14:34] into AI so that we always have something [14:36] to point back to uh as as far as [14:39] copyrightiting. continuing to copyright [14:41] the the future work. [14:43] >> Yeah. And something I just want to plug [14:44] in is that, you know, it's also really [14:46] important to document this human [14:48] creativity. Um because when you go to [14:50] register for copyright, you're going to [14:52] have to show that there's human [14:53] authorship. Now, you don't necessarily [14:54] want to give away because you don't want [14:55] to show them what, for example, the [14:57] original input and what you changed [14:58] exactly. But you do need to be able to [15:00] describe that. So, it's really important [15:02] to document all those things. So, you [15:05] know, although there is concern about [15:06] being able to obtain copyright [15:08] registration for an AI generated work, [15:11] uh obviously we were granted a copyright [15:13] for critters, which was the first one, [15:15] uh ever, which is pretty exciting. Uh [15:18] how did we do it, Lisa? Thank you. [15:21] >> Yes. [15:22] So, you know, it yes, we were excited to [15:25] get copyright registration, but one [15:27] thing I want to point out is that I was [15:29] never terribly concerned about getting [15:31] copyright registration for the film as a [15:33] whole because there were so many human [15:36] elements, the human script and all the [15:38] modifications and and other work that [15:40] was done. Um the thing that we were [15:43] concerned the most about was wanting to [15:46] maximize that protection and not wanting [15:49] to exclude the concept of protection for [15:51] those AI generated elements because I [15:54] said we know the copyright office takes [15:55] that view and they weren't going to you [15:58] know agree to register that and when you [16:00] file a copyright you need to identify [16:02] the AI generated elements explain the [16:05] human authorship and then you're [16:07] supposed to disclaim anything that isn't [16:09] protectable. So our dilemma was you know [16:11] we didn't want to have a long drawn out [16:13] process with the copyright office but we [16:15] also didn't want to con concede that [16:17] view which you know we didn't agree with [16:20] uh because I do think there are good [16:21] arguments that at least with detailed [16:23] text prompts that some of even that [16:26] initial AI generated content is [16:28] protectable. Um so what we decided to do [16:31] was took a bit of a creative approach. [16:33] You know, first of all, we did describe [16:34] all the human authorship and so we [16:36] wanted to make sure that was clearly [16:38] covered. But when it came time to dis [16:40] make a disclaimer unlike a lot of the [16:43] other because there are we think this is [16:44] the first we're trying to come first AI [16:46] generated film uh that at least AI [16:49] dominant film but you know there were [16:51] other works that were registered um and [16:55] they disclaimed they just disclaimed any [16:56] AI generated elements but what we did [16:59] was we disclaimed the AI generated [17:01] elements to the extent not [17:03] copyrightable. [17:05] So, we kind of left and weren't really [17:06] sure how they were going to react to [17:08] that. But the idea was to preserve our [17:09] argument so that if the law changes [17:12] because once you disclaim it, you kind [17:13] of can't go back. Uh, but if the law [17:15] does develop in a way that supports [17:17] that, we didn't disclaim it. And also to [17:19] leave some ambiguity for people who [17:21] might think they want to infringe [17:22] because we've, you know, we haven't [17:23] clearly disclaimed that. Uh, and again, [17:26] we weren't sure how they would react, [17:27] but they accepted the application [17:29] actually without comment. In a lot of [17:31] these situations, you know, there've had [17:32] to be a lot of back and forth, but they [17:34] accepted it outright. So, that was kind [17:36] of exciting. [17:37] >> Can we talk about infringement? [17:40] >> We can. [17:40] >> Okay. So, I I'll just, you know, kick it [17:43] off to you, but another legal issue that [17:45] comes up is the concern about whether AI [17:48] generated content can inadvertently [17:50] infringe on some third party rights. So, [17:52] what practical steps can filmmakers take [17:54] to mitigate legal risks when integrating [17:57] AI into their projects, Lisa? Yeah, [17:59] because this is another, you know, big [18:01] area of risk. People are concerned that [18:03] if they use AI in their films that that [18:05] somebody's going to come after them. And [18:07] that's because, as I'm sure many of you [18:09] aware, there have been at this point [18:10] over 40 lawsuits filed in the US on [18:13] copyright uh infringement issues, [18:16] although these have been consolidated. [18:18] But a couple things to remember. One is [18:20] that so far these suits have all been [18:22] against AI tool companies. there haven't [18:24] been suits uh brought against uh AI [18:28] users. That's not to say that there [18:30] won't be suits, but so far that hasn't [18:32] been the case. But I think that that [18:35] especially in the area of AI film [18:37] making, uh but really with any uh [18:39] generative AI process, there's a lot [18:41] that companies can do to minimize that [18:43] risk. Um and I just want to kind of [18:46] clarify that there's two different [18:47] buckets of risks and issues with [18:49] infringement. One has to do with the [18:50] training process. So you probably hear a [18:53] lot about this of, you know, some [18:54] company, you know, was copying, you [18:56] know, large amounts of somebody's [18:57] property to train their models. Um, and [19:00] that's an issue mostly for the tool [19:02] companies or for companies that train [19:04] their own models. But most filmmakers [19:06] that I work with are using existing [19:07] tools. They're not doing training unless [19:09] they're, you know, fairly sophisticated. [19:11] Um so from a user standpoint the biggest [19:14] issue is output and whether the output [19:16] that's generated is substantially [19:18] similar to other data in that other [19:22] content in that training data. Um, so [19:25] while there is a risk though that what [19:27] you can do because these situations come [19:29] up most often when you're doing [19:31] something like having a prompt that [19:33] says, you know, create X in the style of [19:36] and you're trying to emulate someone [19:38] else's content because if you do that, [19:41] the risk, even though I should note [19:43] style is not protectable [19:45] um, in the US, but if you have a prompt [19:49] like that, you're more likely to have a [19:51] risk of generating content that might [19:53] somebody might think is too similar. So, [19:56] you know, that's one thing that you can [19:58] do very easily to try to uh reduce that [20:00] risk. Um, the other thing that I would [20:03] mention is back to that notion of those [20:06] expressive prompts because if you have a [20:09] prompt and you start with sketches like [20:11] what what Nick was talking about, uh, [20:13] the chances that the output is going to [20:16] reflect something some other content is [20:19] less likely because those sketches are [20:21] going to much heav more heavily [20:22] influence the what the output is. Um, so [20:26] you know there are a few tips that we [20:28] typically run through. So, one of them [20:29] again is don't use prompts that are [20:31] designed to elicit something similar. [20:34] Um, and then think about the tool [20:37] companies that you're choosing because [20:39] it does make a difference. Um, now first [20:41] of all, there are companies out there [20:43] that have tools that only train on [20:45] public domain or licensed materials. [20:48] Now, I'm not saying companies need to do [20:50] that because I do believe that on the [20:52] training side that there are very good [20:53] arguments for fair use. But for [20:55] companies that are not comfortable, that [20:57] can reduce risk. But the other thing to [20:59] note is that there's a variety of [21:01] different approaches by AI tool [21:03] companies on how they train their [21:05] models, how they filter things, and that [21:07] have a much higher or lower risk of [21:09] generating content that could end up [21:12] being somewhat similar. And we don't [21:14] have time to get into the details of how [21:16] that happens, but I think making careful [21:18] choices about the companies you pick can [21:21] reduce risk. Uh and finally, you know, [21:23] many of these companies do have [21:25] indemnities, at least if you have an [21:27] enterprise agreement. So, you really [21:29] want to look at that licensing um and [21:32] and figure out whether you have um you [21:35] know, the ability to um you know, to [21:38] have an indemnity and what the limits of [21:40] that are. I will say as a creator from [21:43] the creator side I've always thought [21:45] that you know the tool providing [21:47] companies should be the ones taking [21:50] responsibility for clearing the con you [21:53] know it unless I'm trying to illicitly [21:56] make like Mickey Mouse and try to [21:58] pretend like I came up with Mickey Mouse [22:00] I think if I'm using safe words and safe [22:03] prompts and inputting my own imagery you [22:06] know I I'm happy that there's an [22:08] indemnity now that a lot of the tool [22:10] companies are offering I I think it [22:11] started with Adobe with Firefly and you [22:13] know now Freepic has them and and Yeah. [22:15] It's the enterprise deals basically. [22:17] >> Yeah. And most of them do because [22:18] there's a distinction between problems [22:20] that arise because of how it's trained [22:22] and problems that arise between how [22:24] somebody prompts and tries to generate [22:26] the content. Um, but Nick, in our last [22:29] little bit, I really wanted to ask you [22:31] for, you know, what is the advice that [22:33] you would give filmmakers who are [22:34] interested in getting into AI, but maybe [22:37] are concerned about the use or just not [22:39] sure really where to start? [22:41] >> I talked to a lot of filmmakers, um, [22:44] both young and more experienced. [22:47] You know, I I talked to a lot of college [22:49] kids that are basically like, hey, I [22:51] don't want to go to school anymore [22:53] because there's no point because [22:54] everything's over. But I I'm I'm much [22:57] more optimistic than that. I think that, [22:59] you know, we're living in an incredible [23:00] time where we're about to see this [23:02] creative renaissance, especially for [23:04] independent cinema. I think, you know, [23:06] the ability for small teams to get [23:07] together and create anything that they [23:10] they can dream up and, you know, get the [23:12] finesse of the traditional knowledge [23:14] with the AI tools and kind of get us to [23:17] the the finish line is a pretty exciting [23:19] time. So, like I wish I was 13 years old [23:21] now growing up with these tools, but I'm [23:23] not. So, it is what it is. But, you [23:26] know, I would say, you know, it's the [23:27] same advice, which is just take the [23:29] tools you have, which now are way more [23:31] than people like me had when I was [23:33] younger, and just make anything you can [23:35] make. And by doing so, you'll also [23:38] realize it's maybe not quite as easy as [23:40] some people make it out to be. And [23:42] there's a lot of headline grabbing going [23:43] on, which is fine. I get it. But you [23:46] know the truth is a lot of the quality [23:48] projects that we put together that we [23:49] work with our clients on take a similar [23:52] amount of time as the other stuff took. [23:54] You know it's there are some [23:55] efficiencies and speed and cost but it's [23:57] not a magic button that you press and [23:59] voila the thing happens. So um my advice [24:02] is to just dive in try to stay a breast [24:05] of all the new things happening which is [24:07] tough because there's new tool [24:08] developments every day but you know do [24:10] your best and tell the story you want to [24:13] tell. [24:15] Yeah. And I think you know a couple [24:17] final takeaways. First of all, one thing [24:18] I meant to mention earlier is that you [24:20] know I've been talking about the rules [24:21] in the US on things like copyright [24:23] ability. The rules are very different [24:24] and you know some countries are similar [24:26] to the US uh and have require human [24:28] authorships. Some don't. Some have had [24:30] for many years going back to the 80s [24:32] this concept of computerenerated work. [24:34] So it's also important to realize that [24:36] you know if you're working globally that [24:38] the laws are going to change. Um but my [24:40] big takeaway is you know the challenge [24:42] is we're in this very uncertain legal [24:44] environment and that can make people you [24:46] know very nervous uh about risks like [24:49] protectability and infringement and [24:51] these are certainly things to take [24:52] seriously and and think about in your [24:55] process and in your planning um but I [24:57] also want to say that I think you know [24:59] particularly in the area of film making [25:00] these aren't risks that are [25:01] uncsurmountable you know if with [25:03] thoughtful ways that of using AI and [25:06] selecting tools and all the things we've [25:08] talked about these are risks that be [25:09] managed So, you know, my advice is just [25:11] is, you know, not to be afraid of [25:13] dabbling in, you know, and and dipping [25:15] your toe into the pool. Um, but just to [25:17] really think about how you're going to [25:19] do it so you do it in a way that can [25:20] really try to minimize that risk. [25:23] >> We got the uh the wave. So, Lisa Oritz [25:26] from Perkins KUI and I'm Nick Cleveroff [25:28] from Native Foreign. Come talk to us [25:29] after. Thank you all.