Film, Music, Anime/Manga × AI — Case Studies (30 cases)¶
For AI characters and conversational NPCs, see ai-characters/case-studies (C01–C14). Disclosure playbook built from these cases → Disclosing AI use Legend: facts carry sources. 💡 marks analysis or hypotheses. Links to domain articles point to the Japanese edition. Japanese edition: ../../domains/case-studies.md
Overview¶
| # | Case | Year | In one line | Issues |
|---|---|---|---|---|
| Film & video | ||||
| F01 | Marvel's Secret Invasion AI opening titles | 2023 | Released during the writers' strike; strong backlash | Timing, disclosure |
| F02 | Netflix's The Eternaut generative AI VFX | 2025 | First major disclosed use in a big production: "not just cheaper, better" | Budget, framing |
| F03 | Disney & Universal v. Midjourney | 2025 | First image-generator lawsuit by major Hollywood studios | Rights in training and output |
| F04 | Disney–OpenAI partnership (Sora) | 2025 | Six months later, the same Disney turned to licensing and investment | "Sue, then license" |
| F05 | AI "actress" Tilly Norwood | 2025 | The actors' union: "not an actor" | Synthetic performers |
| F06 | Lionsgate–Runway partnership | 2024–2026 | Plan to train on its own library stalled, then became an equity stake | Training on own IP, performers' rights |
| F07 | AI images in Late Night with the Devil | 2024 | Three still images were enough for boycott calls | Small uses and disclosure |
| F08 | AI voice correction in The Brutalist | 2025 | AI-corrected pronunciation in an Oscar contender sparked debate | Authenticity of performance |
| F09 | Critterz (OpenAI-backed animated feature) | 2025–2026 | A feature for under $30M in nine months | Budget and schedule |
| F10 | AI de-aging in Here | 2024 | Faces de-aged in real time during the shoot | Combining the actor's own performance with AI |
| Music | ||||
| M01 | "Heart on My Sleeve" (AI Drake / The Weeknd) | 2023 | Unlicensed AI voices went viral and were taken down | Voice rights |
| M02 | Grimes's Elf.Tech | 2023 | The artist opened her own voice with a 50% master-royalty split | Licensing model |
| M03 | The Velvet Sundown | 2025 | A "band" with 1M+ monthly listeners turned out to be AI | Labeling |
| M04 | Deezer's AI music detection and labels | 2025–2026 | Over half of daily uploads are AI; 1–3% of streams | Labeling, fraud |
| M05 | The Beatles, "Now and Then" | 2023–2025 | A song finished with AI stem separation won a Grammy | AI as restoration |
| M06 | Randy Travis, "Where That Came From" | 2024 | A singer who lost his voice released a new song with an AI voice, with his consent | Consent of the artist |
| M07 | Holly Herndon's Holly+ | 2021 | Released a model of her own voice; commercial use decided by a community | Collective voice governance |
| M08 | Spotify's AI policies | 2025 | 75M tracks removed, impersonation banned, AI credits standard | Platform rules |
| M09 | AI artist Xania Monet | 2025 | Suno-made songs charted and won a $3M deal | AI artist with human lyrics |
| M10 | Timbaland's AI artist TaTa | 2025 | Launched as "A-pop" to strong backlash | Marketing AI as a replacement for people |
| Anime & manga | ||||
| A01 | Netflix's The Dog & The Boy | 2023 | AI backgrounds credited as "AI (+Human)"; backlash | Disclosure, respect for creators |
| A02 | Orange (emaqi) AI manga translation | 2024 | From "500 volumes a month by AI" to AI on only ~10% | Translation quality, expectations |
| A03 | Toei Animation's AI plans | 2025 | Backlash to plans in earnings materials; "not currently used" | How to disclose |
| A04 | Twins Hinahima | 2025 | A TV anime using AI as an assist in 95%+ of cuts | Implementing "assistive AI" |
| A05 | Japanese government and CODA requests over Sora 2 | 2025 | Japanese rights holders raised concerns over both training and output | Training vs. output, national stance |
| A06 | Cyberpunk: Peach John | 2023 | A manga drawn with image generation, published in print by Shinchosha | Disclosing the process |
| A07 | Voice actors' "NOMORE unauthorized generative AI" | 2024 | 26 voice actors and industry groups spoke out against unauthorized AI voices | Voice rights |
| A08 | The ChatGPT "Ghibli style" trend | 2025 | Style imitation spread worldwide and reignited debate | Art style |
| A09 | "ChatGPT said" in Crunchyroll subtitles | 2025 | A vendor used AI subtitles in breach of contract | Vendor management |
| A10 | Prime Video's AI English dubs | 2025 | AI dubs of Banana Fish and others were panned and withdrawn | Replacing voice actors, quality |
Film & video¶
F01 Marvel's Secret Invasion AI opening titles (2023)¶
- What happened: In June 2023 the director confirmed the opening sequence was made with AI (produced by Method Studios). Reaction on social media was mostly negative, with calls for a boycott. A storyboard artist posted that it "rubs salt in the wound of every artist and writer" during the WGA strike
- The production's explanation: "AI is one of many tools. It did not replace artists' work"
- Lesson: 💡 The same technology drew more backlash because of timing — in the middle of labor negotiations. Prepare the answer to "whose work did it replace?" before release
- Sources: Deadline / TechCrunch
F02 Netflix's The Eternaut generative AI VFX (2025)¶
- What happened: On the July 2025 earnings call, co-CEO Ted Sarandos disclosed that generative AI was used for a building-collapse scene in the Argentine series The Eternaut — a first for a Netflix original. The in-house VFX team finished it 10 times faster than usual
- How it was framed: "The cost of that VFX just wouldn't have been feasible for a show on that budget"; "an opportunity to make things not just cheaper, but better"; "real people doing real work with better tools"
- Lesson: 💡 Unlike F01, it led with a creative reason — blockbuster-level effects on a smaller budget — and with the human work. The content and order of disclosure change reception
- Sources: TechCrunch / TechRadar
F03 Disney & Universal v. Midjourney (2025)¶
- What happened: In June 2025, Disney and Universal sued the image generator Midjourney for copyright infringement in federal court in Los Angeles, targeting generated images of characters from Star Wars, The Simpsons, Shrek, the Minions and more, and calling it "a quintessential copyright free-rider and a bottomless pit of plagiarism." It was the first suit by major Hollywood studios against an image generator
- Lesson: The dispute is not only about training but also about outputs that reproduce famous characters. Generation services need output-similarity checks (→ ethics-legal, Japanese)
- Sources: CNBC / The Hollywood Reporter
F04 Disney–OpenAI partnership (2025)¶
- What happened: In December 2025, Disney invested $1 billion in OpenAI and signed a three-year license making 200+ animated and creature characters from Disney's brands (Mickey Mouse, Cinderella, Stitch, Ariel, Simba and others) available in the video generator Sora. Talent likenesses and voices are excluded. Disney will also build Disney+ features on OpenAI's API
- Lessons:
- Six months after F03, the same company used both "sue" and "license." Rights holders sue over unlicensed use and open up, for a fee, to partners whose terms fit (the same pattern as Suno/Udio in music → music, Japanese)
- 💡 Excluding human likeness and voice while opening characters appears designed to avoid conflict with the actors' union agreement (→ film-video, Japanese)
- Sources: The Walt Disney Company / CNBC
F05 AI "actress" Tilly Norwood (2025)¶
- What happened: In September 2025, the creator of "Tilly Norwood," a synthetic performer from the AI unit of production company Particle6, said talent agencies were interested, drawing attention. SAG-AFTRA condemned it: "Tilly Norwood is not an actor, it's a character generated by a computer program that was trained on the work of countless professional performers — without permission or compensation." The creator replied that it is "not a replacement for a human being, but a creative work"
- Afterwards: The 2026 actors' union agreement allows synthetic performers only where they add "significant value" compared with a human actor (→ film-video, Japanese)
- Lesson: 💡 Marketing an AI figure as a "replacement for people" collides head-on with labor issues. Positioning it as a character (a fictional being) is more readily accepted (→ ai-characters)
- Sources: Variety / Deadline
F06 Lionsgate–Runway partnership (2024–2026)¶
- What happened: In September 2024, Lionsgate (a library of 20,000+ titles) and video-generation company Runway announced a custom model trained only on Lionsgate's titles. A year later, reports said the plan had stalled: a model trained on the library alone lacked capability, and even its own library raised copyright and performers' ancillary rights issues. Lionsgate later took an equity stake in Runway and expanded to co-producing AI short series based on existing IP
- Lesson: 💡 "It's our own catalog, so we can train freely" is not a given. Rights in the work and performers' rights in likeness and voice are separate, so training on your own IP also requires checking contracts
- Sources: Variety (2024) / TheWrap / Variety (2026)
F07 AI images in Late Night with the Devil (2024)¶
- What happened: In a horror film styled as a 1970s TV show, viewers noticed that three still images shown between segments were AI-generated, and calls for a boycott followed. The directors said they "experimented with AI for three still images" in collaboration with the art team and edited them further; a lead actor said it was sad the backlash undercut the art department's work
- Lesson: 💡 Even a small use draws backlash without prior explanation. The smaller the use, the more likely it is to surface as something "discovered"
- Sources: IndieWire / Forbes
F08 AI voice correction in The Brutalist (2025)¶
- What happened: In a film nominated for 10 Academy Awards, the editor revealed that Respeecher's AI had been used to refine the two leads' Hungarian pronunciation, sparking debate over awards eligibility. The director said the performances were "completely their own," the actors had worked for months with a dialect coach, and AI was used only to make certain vowels and consonants accurate
- Lesson: 💡 The issue was not "was AI used" but whether AI changed what was being judged (the performance). For awards-eligible works, the line between correction and alteration is questioned
- Sources: Variety / Variety (Oscars impact)
F09 Critterz (OpenAI-backed animated feature, 2025–2026)¶
- What happened: In September 2025 it was reported that a short begun with image generation by an OpenAI creative specialist would be expanded into a feature with London's Vertigo Films and others, aiming for a Cannes screening in May 2026. Budget under $30M, nine months of production (major animated features usually cost $100–200M and take three years). Voice actors and human artists take part, and about 30 contributors share profits
- Lesson: 💡 The real effect of generative AI is less "the same thing, cheaper" than making features possible at budgets that never worked before (the same pattern as F02)
- Sources: France 24 / Cartoon Brew
F10 AI de-aging in Here (2024)¶
- What happened: In Robert Zemeckis's film, Metaphysic's generative AI de-aged Tom Hanks and Robin Wright in real time during the shoot. Hanks played five ages, Wright four. No face markers or later compositing were needed
- Lesson: 💡 Uses where the actor performs and AI changes only the appearance do not take work from actors and rarely trigger authenticity debates (F08). They also fit the 2026 union agreement's requirement of consent and compensation
- Sources: The Hollywood Reporter / Variety
Music¶
M01 "Heart on My Sleeve" (2023)¶
- What happened: In April 2023, TikTok user Ghostwriter977 released a song sung by AI voices imitating Drake and The Weeknd. Before removal it had 15M+ views on TikTok and 620,000+ plays on Spotify. After a complaint by Universal Music Group (both artists' label), it was removed from services on April 17
- Lesson: The more an unlicensed voice imitation spreads, the faster it is taken down. In Japan, the Ministry of Justice's August 2026 interpretive guideline states that earning revenue from AI covers with a similar voice can infringe the right of publicity (→ Japan)
- Sources: The FADER / Music Business Worldwide
M02 Grimes's Elf.Tech (2023)¶
- What happened: Right after M01, in May 2023, Grimes released "Elf.Tech," free software that converts vocals into her voice. Commercial releases split master royalties 50/50, administered with distributor TuneCore. Within two weeks, songs using the voice had nearly 1M plays on Spotify
- Lesson: 💡 Against M01's "unlicensed → removed," it showed another option: the artist sets the terms and opens their voice. A reference for voice actors or VTubers releasing an "official voice model" (→ ai-characters P38)
- Sources: NPR / Forbes
M03 The Velvet Sundown (2025)¶
- What happened: In June 2025, a band that entered popular playlists and passed 1M monthly listeners admitted, after days of speculation, to being an AI project. Its bio: "a synthetic music project guided by human creative direction, and composed, voiced and visualized with the support of artificial intelligence," "an artistic provocation" questioning the boundaries of authorship
- Lesson: Listeners had believed they were hearing a human band, and the case fueled calls for streaming services to be required to label AI music
- Sources: Rolling Stone / The FADER
M04 Deezer's AI music detection and labels (2025–2026)¶
- What happened: Deezer publishes the share of daily uploads that are AI-generated: about 10,000 tracks (10%) in January 2025, about 75,000 (44%) in April 2026, and a majority (about 90,000) by June 2026. From June 2025 it became the first streaming service to label AI tracks, and it excludes labeled tracks from algorithmic and editorial playlists
- What the numbers mean: Although most uploads are AI, AI accounts for only 1–3% of streams, 85% of which were detected as fraudulent and not monetized
- Lesson: 💡 The volume of AI output exploded, but the amount listened to did not. In an age of oversupply, value moves to labeling and curation
- Sources: Deezer / TechCrunch
M05 The Beatles, "Now and Then" (2023–2025)¶
- What happened: From a poor-quality 1970s demo left by John Lennon, AI stem separation isolated and cleaned Lennon's voice; the surviving members added their parts. Released in November 2023, it won Best Rock Performance at the 2025 Grammys — the first AI-assisted song to win
- Lesson: 💡 Uses that extract a voice the artist left behind rather than creating one drew almost no backlash. "AI as restoration" is readily accepted
- Sources: Billboard / Loudwire
M06 Randy Travis, "Where That Came From" (2024)¶
- What happened: Country singer Randy Travis, unable to sing after a stroke, released a new song in May 2024 using an AI voice model built from 40+ of his own recordings (1985–2010), layered over a guide vocal by another singer. He and his label were involved; his team calls it the first AI song made with the artist's involvement and consent
- Lesson: 💡 An AI voice with the artist's consent and involvement was received as a story of "getting back what was lost." The technology is nearly identical to M01 (unlicensed); the only difference is consent
- Sources: Rolling Stone / Forbes
M07 Holly Herndon's Holly+ (2021)¶
- What happened: In July 2021, electronic musician Holly Herndon released "Holly+," an AI model of her voice. Anyone can use it for free and publish works, but commercial use is approved by vote of a community (a DAO) of supporters and collaborators
- Lesson: 💡 A precedent, before M02, for "open the artist's voice and govern its use collectively." A reference for official voice models that VTubers or voice actors grow together with fans
- Sources: Scientific American / The FADER
M08 Spotify's AI policies (2025)¶
- What happened: In September 2025, Spotify said it had removed 75M+ "spammy" tracks over the previous 12 months, and announced: (1) a ban on unauthorized voice impersonation, (2) a spam filter against mass uploads and ultra-short tracks, and (3) support for the DDEX industry standard for AI credits (showing in the app whether AI was used for vocals, instrumentation, or post-production)
- Lesson: Platforms are moving not to "ban AI" but to ban impersonation and spam and require AI use to be labeled (the same direction as Deezer in M04)
- Sources: Music Business Worldwide / Variety
M09 AI artist Xania Monet (2025)¶
- What happened: An "AI artist" created by poet Telisha Jones, who writes the lyrics herself, makes the songs with Suno, and uses AI for the visuals. In September 2025 she debuted on Billboard charts (including No. 1 on R&B Digital Song Sales) and, after a bidding war, signed a $3M deal with Hallwood Media
- Lesson: 💡 Artists with a division of labor — human lyrics and concept, AI vocals and visuals — have reached the point of market recognition. There is also criticism that this takes work from human singers
- Sources: Billboard / Music Ally
M10 Timbaland's AI artist TaTa (2025)¶
- What happened: In June 2025, producer Timbaland introduced "TaTa," a fully AI-generated artist from his company Stage Zero, heralding a new genre, "A-pop" (artificial pop). Industry figures and fans objected that it comes at the expense of human artists; the image of a young Asian woman and the treatment of Black culture in the lyrics of an October music video also drew criticism
- Lesson: 💡 Marketing AI as a "replacement for people" draws criticism on both employment and representation (whose image and culture is borrowed) — the same pattern as F05
- Sources: Rolling Stone / NPR
Anime & manga¶
A01 Netflix's The Dog & The Boy (2023)¶
- What happened: A short anime released in January 2023, produced by Netflix, WIT STUDIO and AI company rinna. Backgrounds were made by repeatedly processing hand-drawn concept art with AI and finishing by hand. The background credit read "AI (+Human)," and citing a "labor shortage" as the reason also drew backlash
- Lesson: 💡 The backlash centered less on the technology than on the credit wording and the stated reason. "Labor shortage" is hard to accept in an industry struggling with low pay. Crediting the people involved correctly is the minimum of respect
- Sources: Cartoon Brew / Kotaku
A02 Orange (emaqi) AI manga translation (2024)¶
- What happened: Manga translation startup Orange raised about ¥2.9B (~$20M) from Shogakukan and others in May 2024, promising "up to 500 volumes a month in English." Fans and translators voiced quality concerns, and in September the company explained that AI translation was used for only about 10% of titles on its emaqi store, with the rest handled by about 20 human translators
- Lesson: Translation is where AI has a big effect (→ anime-manga, Japanese), but announcing "massive volume by AI" is read by fans as lower quality. 💡 It is better to state roles up front: "humans supervise," "AI drafts"
- Sources: The Beat / Anime Hunch
A03 Toei Animation's AI plans (2025)¶
- What happened: Its May 2025 earnings materials laid out plans to use AI for auto-generating storyboards, in-betweening, coloring, and turning photos into backgrounds (with investment in and partnership with Preferred Networks). Fans and animators objected to style imitation and job losses; the company removed the images from the materials and added that it "does not currently use AI"
- Lesson: 💡 Even at the planning stage, backlash follows when it becomes visible which process, and whose work, is being replaced. As in A04, frame it as an "assist" and explain how people's work changes
- Sources: Anime Corner / OECD.AI
A04 Twins Hinahima (2025)¶
- What happened: A TV anime (one episode) produced by KaKa Creation and Frontier Works, aired on terrestrial TV on March 28, 2025. AI was used as an "assist tool" in 95%+ of cuts, with hand-drawn finishing. It combined 3D and motion-capture character animation with a hand-drawn cel look, using AI for labor-intensive expressions such as hair movement. It was announced with comments from Yoshikazu Yasuhiko and Macoto Tezka
- Lesson: 💡 Unlike A01 and A03, it set out the idea of "assistive AI" from the start and released it as an experimental work alongside words from well-known creators. Framing and disclosure, more than the use of AI itself, separated reception
- Sources: ITmedia NEWS (Japanese) / Frontier Works (Japanese)
A05 Japanese government and CODA requests over Sora 2 (2025)¶
- What happened: After OpenAI launched the video generator Sora 2 on September 30, 2025, large numbers of videos closely resembling Japanese anime appeared. The Japanese government asked OpenAI not to infringe copyrights. On October 27, the Content Overseas Distribution Association (CODA), whose members include publishers and animation studios, sent a written request asking OpenAI to (1) not train on members' content without permission and (2) respond in good faith to members' claims about generated outputs
- Lesson: Japan has a relatively permissive copyright law for training (Article 30-4), but when outputs reproducing specific works appear en masse, rights holders and the government speak up. Building a business on "Japan is lenient on training" alone is risky (→ Japan)
- Sources: CODA (Japanese) / ITmedia NEWS (Japanese)
A06 Cyberpunk: Peach John (2023)¶
- What happened: A full-color SF manga by writer Rootport, with art drawn using Midjourney. It drew attention in August 2022 after he posted about "having AI draw a manga," and was published in print by Shinchosha on March 9, 2023, with an explanation of the production process and how AI was used in addition to the story
- Lesson: 💡 By disclosing the production process as part of the work, it was hardly seen as "using AI in secret" (contrast with A01 and F07)
- Sources: Shinchosha (Japanese) / ITmedia NEWS (Japanese)
A07 Voice actors' "NOMORE unauthorized generative AI" (2024)¶
- What happened: On October 15, 2024, 26 voice actors including Koichi Yamadera and Yuki Kaji launched "NOMORE無断生成AI," releasing a video saying that "readings, songs, and our voices themselves that we never recorded are being published and sold online." The Japan Actors Union and the Japan Audio Producers Association joined, opposing unauthorized AI generation while calling for rules for coexisting with AI
- Afterwards: In August 2026, the Ministry of Justice's interpretive guideline stated that voices can be protected by the right of publicity (→ Japan)
- Lesson: Japan's voice acting industry took the position not of "no AI" but "no unauthorized use — let's make rules." Games and anime that use AI voices will get there fastest by building consent processes together with the industry
- Sources: ITmedia AI+ (Japanese) / Anime! Anime! (Japanese)
A08 The ChatGPT "Ghibli style" trend (2025)¶
- What happened: When ChatGPT's new image generation launched on March 25, 2025, social media filled with photos redrawn in Studio Ghibli's style. OpenAI's CEO said about 1M new users signed up in an hour. Old footage of Hayao Miyazaki calling an AI demo "an insult to life itself" recirculated, and the ethics of style imitation were debated
- Lesson: Art style itself is hard to protect legally, but mass use of a specific studio's style is received as a question of respect for that brand. Japanese rights holders and the government later made requests to video generators (A05)
- Sources: Variety / Forbes
A09 "ChatGPT said" in Crunchyroll subtitles (2025)¶
- What happened: In July 2025, the German subtitles for the first episode of a new series on Crunchyroll contained the text "ChatGPT said:"; the English subtitles also had many unnatural translations. Crunchyroll said a third-party vendor had used AI subtitles in breach of contract and opened an investigation. Its president had said shortly before that AI was not being considered for the creative process
- Lesson: 💡 Even with a no-AI policy of your own, you cannot fully control how vendors work. Set AI-use rules in contracts and add a step to check deliverables
- Sources: Anime Corner / The A.V. Club
A10 Prime Video's AI English dubs (2025)¶
- What happened: In December 2025, Prime Video added "AI beta" English dubs to anime including Banana Fish and No Game No Life: Zero. Flat delivery even in dramatic scenes was panned as "soulless," and English-language voice actors criticized it. Amazon quickly withdrew the AI dubs — leaving those titles with no English dub at all
- Lesson: 💡 Fans experience dubbing as performance. AI that falls short on quality is judged worse than nothing. Combining AI with human voice actors (building on the actor's own performance, as in F10) is more readily accepted
- Sources: Anime News Network / Engadget
Cross-cutting lessons¶
| # | Lesson | Cases |
|---|---|---|
| 1 | What separates backlash is less the technology than the content, timing, and credits of disclosure. Leading with the human work and a creative reason makes acceptance likelier | F01, F02, A01, A03, A04 |
| 2 | Rights holders are moving to "sue over unlicensed use, license when terms fit" — in film (Disney) and music (UMG, Warner) alike | F03, F04, M01, M02, M07, M08 |
| 3 | Voice and human likeness are increasingly treated separately. Characters may be opened while actors' likenesses and voices are excluded | F04, F05, M02 |
| 4 | The volume of AI output exploded, but what is heard or watched did not. Value moves to labeling and curation | M03, M04 |
| 5 | Even in Japan, rights holders and the government act when outputs resemble existing works | A05 |
| 6 | 💡 "Massive and cheap with AI" is read as lower quality. State roles up front: "assist," "human-supervised" | A02, A03, A04, A10 |
| 7 | AI with the person's consent or own performance is readily accepted. The same technology is judged in opposite ways depending on consent | M01, M05, M06, F10 |
| 8 | 💡 AI's real effect is less "the same thing, cheaper" than "scale and expression that were not viable before": blockbuster VFX on small budgets, a feature in nine months | F02, F09 |
| 9 | AI figures marketed as "replacements for people" are criticized on both employment and representation | F05, M10 |
| 10 | Even with vendors or your own IP, AI-use rights and rules must be checked separately | F06, A09 |