Blendpipe homeSend a brief
Menu

Can you copyright an ad made with AI? What the Copyright Office protects

Can AI generated content be copyrighted? In an ad, only the human-authored parts: footage, animation, 3D, the edit and its arrangement. What to disclaim.

Reader question
Which parts of a commercial made partly with generative AI can the brand register and enforce as its own copyright?
Cropped hands fit a flat drawn steel-blue stone into a real dry-stone wall, its outline matching the gap the surrounding real stones leave.

Can AI generated content be copyrighted? In a commercial, the brand can register and enforce the parts people made: the script, the filmed footage, hand-built animation and 3D, the edit, and the creative selection and arrangement of everything on screen. Material a generative model produced from prompts is not protected; if it is more than de minimis it must be disclaimed in a registration, and the brand cannot use copyright to stop others copying it. AI used as a tool on human work, such as rotoscoping, cleanup or upscaling, does not reduce protection.

That is the position of the U.S. Copyright Office in its registration guidance (effective March 16, 2023) and Part 2 of its report on AI (January 29, 2025), backed by the D.C. Circuit’s ruling that a machine cannot be an author. This article applies those documents to a mixed-media spot, layer by layer. It describes what the Office and the courts have said, as of November 2026; it is not legal advice, and the brand’s counsel decides what to register and how.

Can AI-generated content be copyrighted? The rule in four documents

Copyright protects only material that a human author created. AI-generated material is protected only where a human determined its expressive elements, and the human-authored parts of a mixed work stay protected even when the work also contains AI output. Four primary documents set this out.

The registration guidance, March 16, 2023

The Office’s registration guidance, published at 88 Fed. Reg. 16190, states that “copyright can protect only material that is the product of human creativity.” When an AI system “receives solely a prompt” and produces complex work, the traditional elements of authorship are determined by the technology. That material “is not protected by copyright and must be disclaimed in a registration application.”

The same guidance says applicants “have a duty to disclose the inclusion of AI-generated content” and must give “a brief explanation of the human author’s contributions.” It also says that tools are not the problem: “a visual artist who uses Adobe Photoshop to edit an image remains the author of the modified image.”

Part 2 of the AI report, January 29, 2025

Part 2: Copyrightability is the Office’s fullest analysis. Its conclusions, quoted from the report:

  • “The use of AI tools to assist rather than stand in for human creativity does not affect the availability of copyright protection for the output.”
  • “Copyright protects the original expression in a work created by a human author, even if the work also includes AI-generated material.”
  • “Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements.”
  • “Based on the functioning of current generally available technology, prompts do not alone provide sufficient control.”
  • Human authors are entitled to copyright in their work “perceptible in AI-generated outputs,” and in “the creative selection, coordination, or arrangement of material in the outputs, or creative modifications of the outputs.”

The report also concludes that these questions “can be resolved pursuant to existing law, without the need for legislative change.” Every case, it says, turns on its own facts.

Thaler v. Perlmutter, March 18, 2025

In Thaler v. Perlmutter, the D.C. Circuit affirmed the refusal to register an image whose application named an AI system, the “Creativity Machine,” as its sole author. The court held that “the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being.” It added that the rule “does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence”; the author just has to be the person, not the machine.

The Supreme Court denied review on March 2, 2026 (Docket No. 25-449), so the appeals court decision stands.

The Office’s registration decisions

The Office lists its decisions on AI material on its Copyright and AI page. Two matter most for advertising. In Zarya of the Dawn (February 21, 2023), a comic book’s text and its arrangement were registered while its Midjourney images were excluded. In Théâtre D’opéra Spatial (September 5, 2023), the Review Board refused an image that kept a Midjourney output “in substantial form” because the applicant would not disclaim it. Both are discussed below.

Why prompts alone do not make you the author

The Office treats a prompt as an instruction that conveys ideas, and ideas are not protected. However detailed the prompt and however many times it is revised, the model decides how the instruction becomes pixels or sound, so the person prompting is not the author of that output.

Part 2 says that “prompts essentially function as instructions that convey unprotectible ideas,” and that highly detailed prompts “do not control how the AI system processes them.” It also addresses iteration directly. “Repeatedly revising prompts does not change this analysis,” because effort is not authorship and each revision is like “‘re-rolling’ the dice.” The report compares prompting to commissioning an artist, then notes the key difference: a hiring party can oversee and understand a human artist’s contributions in a way a prompter cannot.

The Colorado case that tested this

The best-known test began in Colorado. Jason M. Allen’s image Théâtre D’opéra Spatial won the 2022 Colorado State Fair’s annual fine art competition, drawing national attention, the Review Board notes, as the first AI-generated image to win it (Review Board decision). Allen told the Office he “input numerous revisions and text prompts at least 624 times” in Midjourney, then edited the result in Adobe Photoshop and upscaled it with Gigapixel AI.

At the first reconsideration, the Office accepted that his Photoshop “visual edits” contained enough original authorship to register; the Board later left that question open for lack of information. Either way, the Office required him to exclude the features generated by Midjourney and Gigapixel AI, and when he declined, it refused the whole claim. The Board noted that Allen himself conceded the upscaler “doesn’t introduce new, original elements.” Allen then sued: Part 2 cites Allen v. Perlmutter, No. 1:24-cv-2665, filed in the U.S. District Court for the District of Colorado on September 26, 2024. We did not find a decision in that case when we checked in November 2026, so watch the docket.

For a producer, the lesson is practical. 624 prompts produced a record of effort, not a record of authorship. What the Office accepted was the hands-on editing, and only to the extent it was separable from the generated image.

Assist or stand in: the line that decides each AI step

The question for every AI step in a production is whether the tool assisted a person’s expression or replaced it. Assistive uses, such as rotoscoping, cleanup, color correction and upscaling of footage people shot, leave protection where it was. Generative uses, where the system chooses the expressive content, need further analysis and usually a disclaimer.

Part 2 records the tasks commenters raised, which will sound familiar to anyone in post-production: “aging” or “de-aging” actors, “removing unwanted objects or crowds from a scene,” and, in its notes, color correction, detail sharpening, de-blurring and rotoscoping. Commenters argued these uses should not affect protection. The Office agrees “that there is an important distinction between using AI as a tool to assist in the creation of works and using AI as a stand-in for human creativity,” and adds that this “depends on how the system is being used, not on its inherent characteristics.”

A three-question test for each step

Before an AI step is approved, ask three questions and write the answers into the plan:

  1. Did a person make the material the tool works on? Filmed plates, drawn frames and built models are human-authored before the tool touches them.
  2. Does the tool add expressive content that no person chose? Removing a wire or tracing a matte adds none; generating a sky, a face or a melody does.
  3. Does the result appear in the delivered spot? If not, the step does not affect the finished work at all.

A step that answers yes, no and yes is assistive. A step that answers yes to the second question produces generated material, which is disclaimed if it is more than de minimis, unless a person’s perceptible input or later edits supply the authorship.

Brainstorming that never reaches the frame

AI used to brainstorm or outline, where a person references the output but does not incorporate it, “should not affect the copyrightability of the resulting human-authored work,” according to Part 2. In production terms, that covers mood frames, AI previs and rough boards that guide a shoot but do not appear in the delivered spot.

A voice example from the report

Part 2 describes a sound recording by the country artist Randy Travis, who has limited speech function after a stroke. The track was built from a recording of a human voice, using an AI vocal model “as a tool” to realize the sounds the human team wanted. Because AI was used as a tool, “not to generate expression,” the Office registered the work.

The contrast for advertising is a synthetic voice that reads a script with no human performance behind it. The script remains human-authored text. The generated performance is a different question, and it also raises consent issues if the voice resembles a real person, covered in our guide to digital replicas and AI voices in ads.

Three ways human authorship survives AI output

Even where a model generated material, three kinds of human contribution can be protected: a person’s own work that stays perceptible in the output, creative modifications to the output, and the creative selection and arrangement of material. Each protects only the human part, never the generated material on its own.

1. Expressive inputs you made yourself

When a person feeds their own copyrightable work into a model and that work remains visible in the result, they author at least that portion. Part 2’s example is Rose Enigma: a hand-drawn illustration used as an input, with a prompt asking for a photorealistic cyborg woman with roses. The Office registered it with an annotation limiting the claim to “unaltered human pictorial authorship that is clearly perceptible in the deposit,” and the applicant disclaimed the AI-rendered nose, lips, rosebuds and lighting.

For advertising, the equivalent is a designer’s own style frame, storyboard or product drawing used as the input to an AI-assisted step. Keep those inputs, because they are the evidence.

2. Modifications that rise to authorship

A person who edits AI output enough can claim the edits. The threshold is real, though. In Zarya of the Dawn, the applicant described modifying a character’s lips and mouth; the Office found the changes “too minor and imperceptible to supply the necessary creativity” (Zarya letter). In the Allen matter, by contrast, the Office accepted at the first reconsideration that Photoshop “visual edits” contained enough original authorship.

Part 2 notes that tools which let a user select a region and regenerate it, such as Midjourney’s “Vary Region,” can give more control than prompts alone. Whether that control reaches the originality standard “will depend on a case-by-case determination.”

3. Selection, coordination and arrangement

The Office registered Zarya of the Dawn’s text and “the selection, coordination, and arrangement of the Work’s written and visual elements,” while excluding the images. Part 2 makes the point that matters most for film and advertising: “a film that includes AI-generated special effects or background artwork is copyrightable, even if the AI effects and artwork separately are not.”

There is a floor. Part 2’s notes say the Office generally will not register a compilation of “only two or three elements,” because that selection is de minimis. A 30-second spot with eight shots, a voiceover, music, supers and an end card is well past that floor, but a single generated image with a logo on it may not be.

A mixed-media spot, layer by layer

A mixed-media commercial is many works in one: a script, footage, animation, 3D, music, voice and the edit that arranges them. Each layer has its own author and its own answer, so treat the spot as a stack rather than one file.

Who made each second of the fictional Orlenna spotA timeline of a 30-second spot with eight shots and four tracks. Live action covers 0 to 7 seconds and 17 to 24 seconds, 14 seconds in all. 2D animation covers 4 to 12, 17 to 20 and 24 to 27 seconds, 14 seconds. 3D covers 12 to 20 and 27 to 30 seconds, 11 seconds. AI-assisted steps cover 0 to 4 seconds (cleanup) and 17 to 20 seconds (rotoscoping), 7 seconds, both applied to human-made footage. Live action 2D animation 3D AI-assisted 0 4 7 12 17 20 24 27 30 s 1 2 3 4 5 6 7 8 cleanup roto Shot
Example (fictional): the Orlenna Ridge Bottle hero spot, an invented brand and product, 30 seconds at 24 fps. Live action 14 s, 2D animation 14 s, 3D 11 s, AI-assisted 7 s; a shot can use several media. Source: the site's fictional example campaign; legal reading from U.S. Copyright Office, Part 2 report (January 29, 2025).

The table applies the Office’s documents to each layer of a typical mixed-media spot. “Protected” means the Office’s documents point to human authorship; the final call on any registration is the Office’s, and on any dispute, a court’s.

Layer Who or what made it Protected? What the record should show
Script, supers, voiceover copy Writers Yes, as text Drafts with dates; who wrote what
Voiceover performance A human performer Yes, in the sound recording Session files and the performer agreement
Filmed footage Director and camera crew Yes Camera originals, shot list
2D animation and kinetic type Animators and designers Yes Project files, style frames
3D models, lighting and renders 3D artists Yes Scene files, render passes
AI-assisted cleanup and rotoscoping People using AI tools on human footage Yes; the Office calls this assistive use Before and after plates, reviewer’s name
AI-generated previs and mood frames A model, from prompts Not needed: not in the delivered spot Mark them “planning only”
Edit, timing, grade, mix Editor, colorist, mixer Yes, including selection and arrangement Edit decision lists, versions
Generated background plate A model, from prompts No; disclaim if more than de minimis Prompt log, what a person changed afterward
Generated product or person shot A model, from prompts No, apart from perceptible human inputs or edits Inputs used, edits made, disclosure decision
Synthetic voice reading the script A model The script yes; the generated performance doubtful Voice tool terms, consent if it imitates anyone
Generated music A model Neither the composition nor the recording, unless a person authored parts Tool terms; human-written parts identified
Licensed music and stock Their owners Theirs, not the brand’s Licenses on the rights sheet

Two layers are often misread. Music is two works: the Office’s Circular 56A explains that a sound recording and the musical work it embodies are “two separate works for copyright purposes,” so a generated track can lack a human author on both counts. And a licensed track or stock shot does not become the brand’s because it sits in the brand’s spot; it stays on the rights sheet with its own term.

Example (fictional). In the Orlenna plan (an invented brand, not a client), every second of the spot is filmed, drawn or built by people. AI touches 7 seconds: cleanup removes a power line in shot 1 (0 to 4 s) and rotoscoping makes the hand matte in shot 5 (17 to 20 s). At 24 fps that is 4 x 24 = 96 frames plus 3 x 24 = 72 frames, so 168 of the spot’s 30 x 24 = 720 frames, or 23.3%. All 168 are human footage processed by tools, the kind of use Part 2 calls assistive, so nothing in the spot is planned as generated material.

Now suppose the client asks to replace the filmed dawn sky in shot 1 with a generated plate. Those 96 frames would then carry generated imagery. The plan would log the prompts, note what the compositor changed, and tell the client that the plate would be disclaimed in any registration, while the rest of the spot and its arrangement stay protected.

The percentage is a planning number, not a legal test. The Office does not register or refuse by share of frames; it asks whether the generated material is more than de minimis and what the human contributed. The count is useful because it makes the disclaimer, if any, easy to describe.

Six common AI steps in ad production, and what each does to ownership

Most AI steps in a commercial fall into a handful of patterns. The answers below apply the Office’s documents to each one; they are planning guidance, and counsel decides any actual registration.

1. Upscaling archive footage

A brand wants to reuse a hero shot from an older campaign, upscaled with an AI tool for a 4K master. Upscaling adds no authorship and removes none. In the Allen matter, the applicant conceded that his upscaler “doesn’t introduce new, original elements,” and Part 2 lists sharpening and de-blurring among the post-production uses commenters raised. The footage keeps whatever status it had, which also means its owner, perhaps a past production company or a stock library, is still its owner. Check the license before the shot goes into the edit.

2. Outpainting a 16:9 shot into a 9:16 version

Generative fill can widen a horizontal shot into a vertical frame instead of cropping it. The added picture is generated material, and the share can be large.

Example (fictional). Take a 3840 x 2160 shot from the Orlenna plan and fit its full width into a 1080 x 1920 vertical frame. The scale is 1080 / 3840 = 0.28125, so the original picture is 2160 x 0.28125 = 607.5 px tall. The model must generate the other 1920 - 607.5 = 1,312.5 px of height: 1,312.5 / 1920 = 68.4% of every vertical frame. Cropping instead keeps a 1,215 x 2,160 area of the source, 32% of the original frame, delivered at 1,080 x 1,920, with no generated pixels at all.

If the brand registers the vertical version, the generated bands are candidates for the “Material Excluded” field. The aspect ratio crop calculator shows what a crop keeps, which is often the cleaner choice for hero shots. Note that disclosure uses a different test: YouTube treats AI “generating or extending a backdrop” behind real action as a minor edit, so a shot can need a disclaimer for registration and no label on upload.

3. A generated crowd or background behind filmed talent

A filmed performer in front of a generated stadium crowd is the case Part 2 describes for film: a work with “AI-generated special effects or background artwork is copyrightable, even if the AI effects and artwork separately are not.” The performance, the camera work and the composite stay protected; the crowd itself does not. If the crowd contains realistic people, the disclosure and likeness questions also apply.

4. AI boards and previs shown to the client

Generated storyboards, mood frames and previs used to agree the plan, then replaced by filmed, drawn or rendered shots, are the brainstorming case. Part 2 says referencing such output without incorporating it “should not affect the copyrightability of the resulting human-authored work.” Mark them “planning only” in the provenance log so nobody mistakes a board for a delivered frame later.

5. Machine translation and synthetic dubbing

The English script stays human-authored text. A Spanish caption file drafted by machine translation and corrected by a fluent person is a mixed case: whether the corrections amount to authorship depends on how much the person changed, and the Office decides that case by case. A Spanish voiceover read by a person is a human performance. A synthetic Spanish voice is generated audio, and if it imitates a real performer, it needs that person’s consent before anything else.

6. A product shot generated from a text prompt

A generated packshot is not protected, however good it looks, and it may not match the real product’s shape, color or label. A product modeled in 3D from the manufacturer’s drawings and rendered by a person is protected human work, can be re-rendered for every version, and shows the actual product. For the shot a brand repeats across a whole campaign, that difference matters twice.

How to register a commercial that contains AI material

Use the Standard Application, describe the human authorship in the “Author Created” field, and exclude AI-generated material that is more than de minimis under “Material Excluded.” If you are unsure, say the work contains AI material and the Office will follow up.

Application fieldWhat the guidance says goes thereExample (fictional) wording for a spot with one generated plate
Application typeThe Standard ApplicationStandard Application, motion picture
AuthorThe human author or authors; never the AI tool or its makerThe brand, as employer for hire, or the named people, as counsel advises
Author CreatedA brief statement of the human authorship; for arrangements, "Selection, coordination, and arrangement of [human content] created by the author and [AI content] generated by artificial intelligence""Script, cinematography, 2D animation, 3D animation, editing, and selection, coordination and arrangement of human-authored footage and animation and a background plate generated by artificial intelligence"
Limitation of the Claim, Material Excluded, OtherAI-generated content that is more than de minimis: "[description of content] generated by artificial intelligence""Sky background plate in seconds 0 to 4 generated by artificial intelligence"
Note to COOptional additional information"AI-assisted rotoscoping and cleanup were applied to human-filmed footage"
If unsureA general statement that the work contains AI-generated material; the Office contacts the applicant"This work contains AI-generated material"
Field names and quoted guidance from the U.S. Copyright Office registration guidance, 88 Fed. Reg. 16190 (March 16, 2023). The example wording is fictional and illustrative; the claimant's counsel decides the actual wording.

The guidance adds that “in some cases, the use of an AI tool will not raise questions about human authorship, and the Office will explain that nothing needs to be disclaimed.” A spot like the fictional Orlenna plan, where AI only cleaned and matted human footage, may fall in that group; counsel decides whether to mention it.

Correcting an application that left AI out

For a pending application, the guidance says to contact the Office’s Public Information Office so a note is added to the record. For a registration already issued, file a supplementary registration that describes the human authorship and disclaims the AI material. Skipping this has costs: the Office “may take steps to cancel the registration,” and a court may disregard a registration if the applicant knowingly gave inaccurate information that would have led to refusal (17 U.S.C. 411(b)). Zarya of the Dawn shows the first outcome: the original certificate was cancelled and replaced with a narrower one.

Why register at all

For a U.S. work, no infringement suit can begin until the claim has been registered or refused (17 U.S.C. 411(a)). Statutory damages and attorney’s fees are generally unavailable for infringement that starts after publication and before registration, unless registration is made within three months after first publication (17 U.S.C. 412). For a campaign that launches on a fixed date, that three-month window is a calendar item, not an afterthought.

Example (fictional). Suppose the Orlenna spot first runs on March 1, 2027, the date the site’s fictional usage example uses. Three months later is June 1, 2027. If Orlenna’s counsel wants statutory damages available for any infringement that starts after launch, the registration needs an effective date by then, which means the provenance log, the “Author Created” statement and any disclaimer must be ready within weeks of launch. Counsel confirms what counts as first publication for an ad.

What the Office has not settled

Some questions remain open as of November 2026, and a plan should not assume an answer:

  • How much editing is enough. The Office accepted Photoshop edits in one matter and rejected lip edits as “too minor” in another. Region editing tools are, in Part 2’s words, a “case-by-case determination.”
  • The Colorado case. Allen v. Perlmutter was pending when we checked; a decision could refine where the line sits.
  • Updated guidance. Part 2 says the Office will issue “additional registration guidance and an update to the Compendium.” Its AI page, checked November 22, 2026, still lists the March 2023 statement as its registration guidance.
  • Training. Part 3, on training models with copyrighted works, was released as a pre-publication version on May 9, 2025; the Office says a final version will follow. It concerns the tool’s inputs, not your ownership of the spot.

What a competitor could reuse, and what it could not

Generated material that the brand disclaimed, or could not claim, gives the brand no copyright to enforce, so a competitor copying that material alone would not infringe the brand’s copyright. The human-authored footage, animation, 3D and the spot’s arrangement remain protected, and copying those is a different matter.

Where the exposure sits

The risk depends on which elements are generated. A generated background behind a filmed product is low stakes: copying the background gets a competitor little. A generated mascot, hero character or signature visual is different, because the brand may want to stop others using exactly that, and copyright may not help.

As a rule of thumb for planning, keep the elements the brand will want to defend human-authored:

  1. Characters and mascots: designed and animated by people, with AI help limited to assistive steps.
  2. Signature transitions and visual devices: built by hand in animation or 3D.
  3. Packshots and product renders: modeled from the product drawings and rendered, not generated.
  4. Music the brand will reuse across campaigns: composed and recorded by people, or licensed with clear terms.
  5. Taglines and supers: written by people, which is already standard.

Example (fictional). Suppose Orlenna had used a generated sky plate in shot 1 and disclaimed it. A competitor that lifted only that sky for its own ad would be copying material Orlenna has no copyright in. A competitor that copied shot 1 as a whole, with the filmed runner, the framing, the timing and the cut into shot 2, would be copying human authorship and the arrangement, which the registration covers. The disclaimed plate narrows what Orlenna can enforce; it does not open the spot to copying.

Other protections are separate questions

Copyright is not the only body of law. Trademark, unfair competition and contract terms work differently and fall outside this article; counsel decides whether any of them covers a given element.

Tool terms are a frequent source of confusion. A generative tool’s terms may assign the user whatever rights exist in the output, but a contract cannot create a copyright the law does not recognize. What the main tools’ terms say about ownership, commercial use and training is covered in our guide to AI video tool terms.

Who owns the human-authored parts

The human-authored parts belong to whoever the law and the contracts say: the individual creators, their employer under the work made for hire rules, or whoever receives an assignment. AI does not enter that chain, because it cannot be an author, a co-author or a party to a contract.

The D.C. Circuit’s Thaler opinion summarizes the work made for hire rule: the Copyright Act allows “the employer or other person for whom the work was prepared” to be “considered the author,” and such a copyright lasts 95 years from first publication or 120 years from creation, whichever expires first. Part 2 adds in its notes that AI systems “cannot produce joint works or works made for hire because they are not ‘authors’” and “cannot enter into binding contracts.”

For commissioned work, the statute is specific. Under 17 U.S.C. 101, a work “specially ordered or commissioned” counts as made for hire only if it falls in a listed category, which includes use “as a part of a motion picture or other audiovisual work,” and only “if the parties expressly agree in a written instrument signed by them.” A commercial is an audiovisual work, so the written agreement with each freelancer, studio or agency is what decides whether the brand is the author from the start or needs an assignment.

How footage, project files and raw material pass from a production company to a brand is a contract question. It is covered in who owns commercial footage, and the delivery side is listed on deliverables.

Production records that make the claim provable

A registration statement is only as good as the records behind it. The Office asks applicants to describe the human contribution and exclude the AI material, so the production needs to know, shot by shot, which is which.

What to keep

  1. A provenance log listing every AI-assisted step: the shot, the tool category, what the step did and the person who reviewed it.
  2. Human inputs to any AI step: style frames, drawings, boards or plates a person made.
  3. Before and after files for assistive steps such as cleanup and rotoscoping, so the human footage underneath is visible.
  4. Prompt logs for any generated material, not because prompts create authorship (the Office says they do not) but because they show what the model, not a person, decided.
  5. Edit decision lists and versions that document the selection and arrangement.
  6. Names and roles of the people who wrote, shot, animated, built, edited and approved.

Making the records match the claim

The useful habit is to plan the claim before production. If the brand intends to register the spot, the production plan says so, and every AI step in the plan carries a note: assistive (no effect expected), generated (disclaim if used), or planning only (not in the delivered work).

Example (fictional). Orlenna’s plan lists four AI-assisted steps. Previs frames for shot 1: planning only, never in the delivered spot. Power line cleanup in shot 1: assistive, applied to filmed footage, reviewed frame by frame. Hand matte in shot 5: assistive, finished by a compositor. Spanish caption drafts: text checked and corrected by a fluent reviewer. At handover, the provenance log lists all four with reviewers, and the rights sheet lists the licensed music separately. If Orlenna’s counsel registers the spot, the “Author Created” statement can be written from those two documents, without reconstructing the production from memory.

Decisions to make before an AI step is approved

Ownership is decided at the planning stage, not at registration. Once a generated element is in the cut, the only remaining choices are to disclaim it or replace it.

  1. Name the elements the brand must own (characters, signature visuals, packshots, music) and keep them human-authored.
  2. Classify every AI step as assistive, generated or planning only, and record it in the plan before work starts.
  3. Keep human inputs to any AI step, because perceptible human expression is protectable.
  4. Check each tool’s terms for commercial use, output rights and training on uploads, remembering that terms cannot create copyright.
  5. Check the disclosure side as well. A generated element that raises a copyright question may also need a platform label; see AI disclosure labels on video ads.
  6. Decide on registration early, so the three-month window after first publication is in the launch calendar.
  7. Deliver the records: provenance log, rights sheet and project files as the contract specifies.

Blendpipe is a production studio in Colorado Springs that combines live action, animation, 3D and AI-assisted tools, and its production plans tell the client which parts of the work are human-authored and which steps used AI tools. The policy behind that is on AI practices, and the kinds of AI-assisted steps involved are described on AI-assisted production. To see how a plan records each step, read the fictional example campaign, or send a brief for your own.