{"id":28093,"date":"2025-10-29T16:46:06","date_gmt":"2025-10-29T15:46:06","guid":{"rendered":"https:\/\/explainvisually.co\/?p=28093"},"modified":"2026-01-22T12:36:58","modified_gmt":"2026-01-22T11:36:58","slug":"ai-animation-cost-saver-or-strategic-trap-6-risks","status":"publish","type":"post","link":"https:\/\/explainvisually.co\/en\/ai-animation-cost-saver-or-strategic-trap-6-risks\/","title":{"rendered":"AI animation – cost saver or strategic trap? 6 risks"},"content":{"rendered":"

On LinkedIn and in boardrooms, one question is being asked more and more often:<\/p>\n

\u201cWhy should we pay an animation studio when we can just click it out ourselves-faster and cheaper?\u201d<\/strong><\/p>\n

Every day, we see videos generated by Sora and other AI models that look truly impressive.
\nThe promise is extremely tempting: you press \u201cgenerate\u201d<\/em> and get a ready-made ad.
\nSounds like every marketing director\u2019s dream-someone who must watch both deadlines and budgets.<\/p>\n

The phrase \u201csounds like a dream\u201d<\/em> is key here-because the reality is still far from it.<\/p>\n

In this article, I\u2019m not rejecting technology, nor claiming that AI is useless. It isn\u2019t.
\nI use artificial intelligence daily and know firsthand how much it can streamline processes and even support creative work.
\nBut it still can\u2019t do everything-and it comes with several serious risks.<\/p>\n

Risk #1<\/strong><\/h2>\n

Legal no man\u2019s land (a key issue for corporations)<\/strong><\/h2>\n

You invest in advertising to build unique value-a piece of content that belongs exclusively to you. And here comes the first, enormous problem with AI.<\/p>\n

For something to be protected by copyright, it must be the result of human creativity<\/strong>. The position of global institutions-led by the U.S. Copyright Office, which sets the tone for the market-is clear: works generated 100% by AI are not eligible for copyright protection.<\/strong><\/p>\n

Example:<\/strong> The well-known case of the comic \u201cZarya of the Dawn.\u201d <\/em>Author Kris Kashtanova initially received copyright protection but failed to disclose that the images were created using Midjourney.<\/p>\n

Upon review, the USCO made a landmark decision: it revoked protection for the images, keeping it only for the text and the human-made layout and composition of the panels.<\/p>\n

Even more, the office rejected Kashtanova\u2019s claim that later manual edits qualified the illustrations for protection, stating the modifications were \u201ctoo minor or imperceptible to meet the threshold of human authorship\u201d or that the agency \u201ccould not determine the actual human contribution from the submitted evidence.\u201d<\/p>\n

\"\"<\/p>\n

“Zarya of the Dawn\u201d Kris Kashtanova \/ Midjourney<\/em><\/strong><\/p>\n

Imagine AI generates the perfect logo, brand mascot, or campaign visual for you. The problem? You can\u2019t trademark it. You don\u2019t have exclusive rights.
\nThat means your competitor could legally copy and use the same motif or character tomorrow.<\/p>\n

The risk of \u201cpoisoned data\u201d<\/em> is the flip side. You never have full certainty about what the model was trained on. AI systems were trained on billions of images and frames from the internet-often without licenses or the creators\u2019 consent. There\u2019s a real possibility that your \u201cunique\u201d animation is, in fact, an unconscious copy or collage of copyrighted works belonging to other artists, studios, or corporations (like Disney). Legally, you<\/strong> are responsible. As the publisher, you-not the algorithm-bear the risk of copyright infringement lawsuits.<\/p>\n

Examples of lawsuits that shook the market:<\/p>\n

Getty Images vs. Stability AI<\/strong><\/span><\/h4>\n

One of the world\u2019s largest stock photo agencies sued Stable Diffusion\u2019s creators, Stability AI, for \u201cflagrant infringement\u201d of intellectual property.
\nAllegation: the model was trained on 12 million Getty Images photos-without a license-and generated images that even reproduced Getty\u2019s distinctive watermark.<\/p>\n

\u201cThey\u2019re a bunch of tech fanatics so excited about AI that they ignored every potential danger or problem it could cause,\u201d said Lindsay Lane KC, Getty\u2019s lawyer in the London trial.<\/p>\n

The company\u2019s legal team went further, claiming Stability AI was \u201ccompletely indifferent to what it fed into its training data,\u201d resulting in \u201cour trademark being attached to pornography\u201d and \u201cAI-generated junk.\u201d<\/p>\n

\"\"<\/p>\n

Illustrations attached to the Getty Images lawsuit. On the left – the original, copyrighted Getty Images photograph. On the right – a similar image generated by Stable Diffusion, which even recreated the company\u2019s distorted watermark.<\/em><\/strong><\/p>\n

Artists vs. Stability AI & Midjourney<\/strong><\/span><\/h4>\n

Another high-profile case: a class-action lawsuit filed by artists including Sarah Andersen against Stability AI, Midjourney, and DeviantArt. The accusation: instead of protecting its artists, DeviantArt allegedly supplied their works for model training-and profited from it.<\/p>\n

\"\"<\/p>\n

Artwork by artist Sarah Andersen – source: <\/em>https:\/\/perma.cc\/8AEZ-7B3W<\/em><\/a><\/strong><\/p>\n

\"\"<\/p>\n

Comic generated with the help of AI. \u201cIn this comic, only the text was edited. The drawings are mine (though they\u2019re a bit outdated, as my style has evolved since 2017). I found this typeface to be a convincing imitation of my handwriting. If the text had been fitted into the speech bubbles more elegantly, I would have considered the whole thing almost indistinguishable from my own work,\u201d wrote artist Sarah Andersen in her essay in The New York Times.<\/em><\/strong><\/p>\n

Researchers from UC Berkeley and ETH Zurich demonstrated that popular AI models can reproduce near-identical copies of their training data.
\nWhen they asked one model to generate an image of \u201cAnn Graham Lotz,\u201d it returned her Wikipedia portrait-differing only by minimal noise (tiny random pixel variations).<\/p>\n

Among the reproduced content were copyrighted press photos, film stills, and even corporate logos.<\/p>\n

They also analyzed training data legality: 35% of images had an explicit copyright notice forbidding reuse, and another 61% were likely protected even without such notice. Worse, some datasets contained sensitive material-medical imagery and private photos.<\/p>\n

In short: the risk that your \u201cunique\u201d visual-or animation-is, in reality, plagiarism<\/strong> is both real and measurable.<\/p>\n

If you use AI merely as an assistant, not a creator, the situation improves-but only slightly.<\/strong><\/p>\n

The U.S. Copyright Office states that legal protection depends on the degree of human creative input<\/strong>.<\/p>\n

What doesn\u2019t<\/em> count as creative input? Typing \u201cgenerate me a cat in the style of Van Gogh\u201d<\/em> does not<\/strong> make you an author. You\u2019re just a tool user.<\/p>\n

What can<\/em> count? If an artist takes AI-generated fragments and significantly reworks them-redraws characters, merges elements into a complex collage, performs deep corrections and edits in graphic software – then the final piece<\/strong> may qualify for protection, but only if the human contribution is dominant and substantial.<\/p>\n

Risk #2<\/strong><\/h2>\n

The \u201cblack box\u201d trap<\/strong><\/h2>\n

The term \u201cblack box\u201d perfectly describes AI\u2019s biggest business problem. You feed in a prompt-but have no control<\/strong> over what happens inside. You can\u2019t edit the process. You only get the final result the machine decides to produce.<\/p>\n

That\u2019s why single AI-generated images can look impressive.<\/p>\n

But animation requires many<\/em> consistent, brand-specific frames-and that\u2019s where AI (still) fails.<\/p>\n

AI models excel at generating broad concepts. Type \u201cretro-style businessman in an office animation,\u201d and you\u2019ll get dozens of options.<\/p>\n

The problem starts when your brief is specific:<\/p>\n

    \n
  • You need to show your<\/em> production hall, with your<\/em> assembly line.<\/li>\n
  • Characters must wear your<\/em> uniforms, with the logo in the exact<\/em> brandbook position.<\/li>\n
  • Key elements must match your brand colors.<\/li>\n<\/ul>\n

    AI hallucinates details. It\u2019ll create a<\/em> factory, not your<\/em> factory. A<\/em> uniform, not your<\/em> uniform-and the logo will appear as a blurry blob.<\/p>\n

    For a marketing department that guards brand consistency, such material is a major issue.<\/p>\n