In February 2026, short clips of Spider-Man, Darth Vader and Grogu started circulating online, all of them generated with Seedance 2.0, ByteDance's text-to-video model. Disney's lawyers noticed quickly. A cease and desist letter arrived, other studios sent their own, a Hollywood trade group went public, and ByteDance paused its global rollout. By August an industry agreement was signed, yet the biggest legal question stayed open. If you make AI video for clients, for a channel, or just for fun, the real worry is not who wins the studio fight. It is whether the fallout can reach you. Here is what the letters say, what lawsuit risk looks like in practice, and how to keep your own clips on safe ground.
💡 This article is general information, not legal advice. For a specific project, talk to an intellectual property attorney.
What Disney Actually Sent
The Letter in Plain Words
Disney's outside counsel, David Singer, sent the cease and desist letter to ByteDance on a Friday in mid-February 2026, and press reports date it to February 13. The central accusation is blunt: ByteDance shipped Seedance 2.0 "pre-packaged" with a pirated library of Disney's copyrighted characters from Star Wars, Marvel and other franchises, and treated them like free clip art.
The letter attached sample outputs showing Spider-Man, Darth Vader, Grogu and Peter Griffin. Singer described the behavior as a "virtual smash-and-grab" that is "willful, pervasive, and totally unacceptable." That word, willful, is not casual. Under US copyright law, willful infringement raises the ceiling on statutory damages, so lawyers use it on purpose.

What a Cease and Desist Does
A cease and desist letter is a demand, not a court filing. No judge reviews it, nothing is ordered, and nobody has been found liable. A typical letter contains:
- A description of the rights holder's works and the alleged infringement
- A demand to stop the conduct and remove infringing material
- A request to preserve documents, which is a polite hint that litigation may follow
- A deadline for a written reply
It also builds a paper trail. Once a company has been told in writing that something infringes, continuing looks less like an accident and more like a choice. That is exactly why these letters go out before lawsuits do.
Who Else Joined the Pressure
Disney was far from alone. Motion Picture Association chair Charles Rivkin said on February 10 that the model had "engaged in unauthorized use of U.S. copyrighted works on a massive scale." Paramount sent its own letter stating that its intellectual property was used to train the models behind these tools without consent. Netflix reportedly threatened "immediate litigation" and demanded guardrails. SAG-AFTRA publicly sided with the studios.
The pattern matters more than any single letter: studios, a trade group and a performers' union all pointed at the same product within days of each other.
How ByteDance Responded
Safeguards and a Pause
On February 16, a ByteDance spokesperson said the company respects intellectual property rights and was "taking steps to strengthen current safeguards" against unauthorized use of IP and likeness. The statement did not say what those steps were. In March, Senators Marsha Blackburn and Peter Welch urged ByteDance to shut the platform down and comply with US copyright law. On March 17, ByteDance voluntarily suspended the global rollout, according to reports.
| Date | Event |
|---|
| February 10, 2026 | MPA chair publicly accuses the model of large scale unauthorized use |
| February 13, 2026 | Disney's cease and desist letter reaches ByteDance |
| February 16, 2026 | ByteDance promises stronger safeguards |
| March 2026 | Two US senators demand a shutdown |
| March 17, 2026 | Global rollout paused |
| August 18, 2026 | MPA and ByteDance sign a formal agreement |

The August Agreement
On August 18, 2026, the MPA and ByteDance signed a formal agreement that applies to Seedance, Seedream, and the outputs of TikTok, CapCut and Dreamina. It formalizes output-layer protections: content filters, face-blocking, and C2PA Content Credentials, which are cryptographically signed provenance records attached to generated media.
Here is the catch. The agreement explicitly does not say whether training on copyrighted material was infringement, which was the core allegation in the original letters. It limits what the tool produces from now on. It does not settle the past.
💡 Content Credentials in practice: AI clips can carry a traceable label. Expect platforms and clients to start reading it.
Is a Lawsuit Coming?
Letters Are Not Lawsuits
According to reporting through August 2026, no major studio had filed a federal lawsuit against ByteDance over the model. One reason given is practical: suing a foreign company means serving papers abroad, and service under the Hague Service Convention can take 18 to 24 months. That can change at any time, and a letter can be followed by a complaint months later.
What usually happens after a letter like this? The recipient replies in writing, the lawyers negotiate, and most disputes end in one of three ways: the company changes its product, the parties sign a license, or the rights holder files suit. The August agreement looks like the first outcome. It is a product change with a trade group's blessing, not a verdict.
What Midjourney and MiniMax Show
Studios do sue when they choose to. Disney and Universal sued Midjourney in June 2025, describing it as a "bottomless pit of plagiarism." Warner Bros. filed a similar suit that September, and the two cases were combined in November. Reports of the complaint cite statutory damages of up to $150,000 per image. A court has since suggested mediation.
Video is not exempt. On September 16, 2025, Disney, Universal and Warner Bros. sued MiniMax, the company behind Hailuo AI, in federal court in California. A judge refused to dismiss the case, finding plausible claims of direct and secondary infringement.

Notice who gets named in those complaints: the companies that build and sell the generators, not the individuals typing prompts.
The other side of the coin is licensing. In December 2025, Disney announced a three-year deal with OpenAI: a $1 billion equity investment and licensed use of more than 200 characters from Disney, Pixar, Marvel and Star Wars in Sora and ChatGPT Images. The deal excluded talent likenesses and voices. Disney is not against AI video. It is against unlicensed AI video that borrows its characters for free.

Can Creators Get Sued?
Users Versus Platforms
The honest answer: yes in principle, rarely in practice. Copyright law lets rights holders pursue whoever copies or distributes protected work, and a user who generates and publishes a Darth Vader clip is distributing it. So far the letters have gone to the platform. For creators, the usual fallout is smaller but real: takedown notices, channel strikes, demonetization and suspended accounts.
Here is how that usually plays out. A rights holder files a takedown notice with the hosting platform, the platform removes your clip, and you get a strike. Collect enough strikes and the whole channel can go. You may file a counter-notice if you believe the clip is lawful, but doing so exposes your name and address to the claimant, so it deserves a lawyer's opinion first.
Risk climbs sharply when money enters the picture.
| Scenario | Risk | Why |
|---|
| Original characters in an original setting | Low | Nothing protected is copied |
| Generic archetypes such as a knight or a robot | Low | Ideas and archetypes are free to use |
| A broad genre look or era | Low to medium | Style alone is generally unprotected, but outputs can drift into specific designs |
| Fan clip with a named studio character, posted free | Medium | Direct copy of a protected character, takedown likely |
| The same clip inside a paid ad or on merchandise | High | Commercial use, plus a trademark claim |
| A real actor's face or voice | High | Likeness and publicity rights, union attention |

Trademark and Likeness Rules
Studio characters are protected twice. Copyright protects the characters' expression: designs, stories, voices. Trademark protects names, logos and costumes that tell the public who made something. Real people add a third layer, the right of publicity, which is why performers' unions react so quickly to face and voice clones.
Where Fair Use Stops
Fair use is a defense you argue after someone accuses you. It is not a permission slip. Courts weigh four factors: the purpose of the use, the nature of the original work, how much was taken, and the effect on the market for the original. Parody and commentary have the strongest arguments, but "Spider-Man fights in the rain" with no commentary or critique is a hard sell, because a clip that stands in for official content hurts the market for it. And "an AI made it" is not a defense of any kind: the person who publishes the clip owns the publishing decision.
The Training Data Question
Two separate issues hide inside this fight. One is what goes in: whether training a model on copyrighted films is itself infringement. The other is what comes out: whether a generated clip copies a protected character. Paramount's letter attacked the first. The August agreement only touched the second.
Courts have not settled the first one. Rulings in text-based AI copyright cases during 2025 went in different directions, and no ruling yet addresses video generation at this scale. Nobody can honestly promise how it ends.

💡 Training lawsuits move in years. Your prompt habits change your risk this afternoon.
For a working creator, that is the useful split. You cannot control what a model was trained on, but you fully control what you ask it to produce and what you publish.
Clients are already moving this way. Agencies and brands increasingly ask which tool made a clip, what the prompt was, and whether any protected character or real face appears in it. Showing a clean prompt history and a platform that attaches provenance data answers those questions before they become a contract dispute.
Safer Ways to Make AI Video
Invent Your Own Characters
Describe traits, not names. A distinctive original character gives you something you can own, reuse and defend. Compare these prompts:
| Risky prompt | Safer rewrite |
|---|
| "Darth Vader cooking in a kitchen" | "A retired space marshal in patched brown armor stirring soup in a cramped galley kitchen" |
| "Spider-Man swinging through Tokyo" | "A rooftop courier in a homemade costume vaulting between buildings at dawn" |
| "Baby Yoda sipping soup" | "A small mossy forest spirit sipping tea from a chipped cup" |

Keep Prompts Free of Names
- Leave out character names, franchise names and actor names.
- Do not upload frames, posters or screenshots of protected characters as references.
- Do not upload photos of real people without their permission.
- If a tool lets a protected character through, remember that a missing filter is not a license.
Describing a famous character without naming it does not make it safe. "A masked hero in a red and blue suit with a web pattern, swinging between skyscrapers" points at the same protected design, and a court looks at what the clip shows, not at the words you typed. Change the design itself: new colors, new silhouette, new gear, new backstory.
Save Your Prompt Records
Keep every prompt, seed, setting, source image and date for commercial projects. If a claim ever arrives, a clean record showing independent creation is worth more than any argument made afterward. Also keep any provenance metadata the tool attaches.

How to Use Seedance 2.0 on PicassoIA
Seedance 2.0 on PicassoIA turns a text prompt or a reference image into a finished video with built-in audio. Use it with original subjects, and the settings below keep your workflow clean.
Step by Step
- Open the Seedance 2.0 model page on PicassoIA.
- Write a prompt about an original subject: who it is, where it happens, what moves, and how the camera behaves.
- Choose a duration. The default is 5 seconds, and setting it to -1 lets the model pick the length.
- Set the resolution (480p, 720p, 1080p or 4k) and the aspect ratio (16:9 by default, 9:16 for vertical, or adaptive).
- Leave audio generation on if you want dialogue in double quotes, sound effects and background music.
- Generate, review the result, and adjust the prompt before publishing.
Settings Worth Using
- Reference images: up to 9 images of your own original character keep faces and outfits stable between clips. Never upload protected characters here.
- First and last frame: pin the start and end of a clip to control exactly how it opens and closes.
- Seed: fix it for reproducible results, and write it down with your records.
- Reference audio: up to three clips for lip-sync, but only with voices you have the right to use.
If you want other options, PicassoIA hosts several close alternatives:

💡 Access to a model is not a license to any character. The same rules apply on every tool.
Make Something Original Today
The legal fight between studios and AI video makers will run for years, and the headlines will keep changing. What stays constant is simple: original ideas are yours, borrowed characters are risky, and good records protect you.
The best response is to build something nobody can send a letter about. Open Picasso IA, pick a video or image model, and create your own character, your own world, your own story. Try a quiet rooftop courier, a talking teapot, or a retired marshal with a soup problem. Generate a few clips, change one detail at a time, and see how far an original idea goes.