Most AI video tools allow their output in paid ads, but commercial permission is the only thing their terms reliably give you. “Ownership” usually means the vendor does not claim the output, or assigns whatever rights it has “if any”; many individual plans let the vendor train on your uploads; and legal cover for infringement, where it exists, sits mainly on paid business tiers and, in the terms read here, excludes modified output or trademark claims. Before an AI-generated shot goes into a campaign, check five things in the terms of the exact plan used: the tier, ownership, training on client material, indemnity scope and limits, and likeness and trademark rules. Then write the answers into the rights sheet.
This guide reads the published terms of three vendors of generative video and image tools (Runway, OpenAI and Google), plus Adobe Stock’s note on generated outputs, each dated as checked on November 22, 2026. Naming them is not a recommendation; they are examples of how the clauses are written. It also uses the US Copyright Office’s January 2025 report to separate the right to use an output from owning a copyright in it.
Why the plan tier decides which terms apply
The same vendor often runs two contracts: individual terms for self-serve accounts and separate business terms for enterprise or paid API customers. The answers to every other question in this guide can flip between them, so the first fact to establish is which contract governs the account that generated the shot.
The documents say so in their opening lines:
| Vendor | Individual or self-serve terms | Business terms |
|---|---|---|
| Runway | Terms of Use, last updated October 23, 2026 | Enterprise Services Terms, last updated June 1, 2026, which apply to “organizations purchasing a subscription to Runway Enterprise Services” |
| OpenAI | Terms of Use, effective January 1, 2026 | Services Agreement, effective January 1, 2026, for “APIs, ChatGPT Enterprise, ChatGPT Business” and other business services, plus the Service Terms, updated November 13, 2026 |
| Gemini API Additional Terms, effective March 23, 2026, which treat “Unpaid Services” and “Paid Services” differently | Google Cloud Service Specific Terms, last modified November 18, 2026 |
Three consequences follow for a production. A freelancer’s personal account is governed by individual terms even when the shot is for a client. A free trial or free quota is often treated differently from paid use. And the party called “Customer” in business terms is whoever signed: if the agency holds the account, the brand is not the customer and gets no direct benefit from the vendor’s promises.
Which AI steps in an ad trigger which questions
Not every AI step carries the same risk. A text-to-video generation creates a new picture from a prompt, while an AI-assisted cleanup changes filmed footage the brand already controls; the first raises ownership and infringement questions, the second mostly raises questions about what was uploaded.
| AI step | What usually goes into the tool | Questions that matter most |
|---|---|---|
| Text-to-video or text-to-image generation | A prompt, sometimes style references | Ownership and uniqueness; output indemnity; trademarks in the output |
| Image-to-video from a client reference | Product photos, renders, key art | Training on uploads; ownership; trademark and design rights |
| Video extension or inpainting of a filmed plate | Frames of the client’s footage | Training on uploads; whether the result counts as modified output |
| AI-assisted rotoscoping, cleanup or upscaling | Frames of the client’s footage | Training on uploads; where the frames are processed |
| Voice generation or voice localization | A script, sometimes a voice sample | Voice consent; disclosure duties in the terms; indemnity exclusions |
| Concept frames and previs that never air | Scripts, boards, references | Training on uploads; confidentiality of an unannounced campaign |
Two rows deserve a second look. Previs and concept frames never reach the screen, yet they often carry the most confidential inputs: an unannounced script, a product that has not launched. And cleanup or rotoscoping may run on a cloud service even when it is launched from desktop software, so find out where the frames are processed, not only which application the editor opened.
Commercial use and ownership: what the terms actually grant
Read closely, the terms grant permission to use output commercially and a promise that the vendor will not claim it. None of them can give you a copyright the law does not recognize, and all of them warn that someone else may receive similar output.
What each set of terms says about output
- Runway (Terms of Use, section 4.4): “The Company does not claim ownership of any of your Inputs or Outputs. Subject to your compliance with the Agreement, the Company does not restrict your commercial use of your Outputs.”
- Runway (Enterprise Services Terms): “As between the parties, Customer owns all rights, title, and interest in Customer Content,” which the terms define to include uploaded files and generated outputs.
- OpenAI (Terms of Use): you “retain your ownership rights in Input” and “own the Output,” and OpenAI assigns “all our right, title, and interest, if any, in and to Output.” The Services Agreement (section 4.1) uses the same structure, “to the extent permitted by applicable law.”
- Google (Gemini API Additional Terms): “Google won’t claim ownership over that content.”
- Google Cloud (Service Specific Terms, section 20): “Google does not assert any ownership rights in any new intellectual property created in the Generated Output.”
The phrases “if any” and “to the extent permitted by applicable law” carry the weight. A vendor can assign only the rights it has, and for purely generated material those may be none.
Permission to use is not copyright ownership
The US Copyright Office’s Part 2 report on copyrightability (January 2025) concludes that copyright “does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements,” and that “prompts do not alone provide sufficient control.” It also finds that “the use of AI tools to assist rather than stand in for human creativity does not affect the availability of copyright protection,” and that human authors are entitled to copyright in their expression “perceptible in AI-generated outputs” and in creative selection, arrangement or modification.
So a contract clause saying you “own the Output” settles the matter between you and the vendor. It does not stop a competitor from reusing a frame of a purely generated shot that is not protected. Which parts of a finished spot the brand can register is a question for counsel, covered in Can you copyright an ad made with AI?, and who holds what between brand and production company is covered in who owns the footage, project files and AI elements.
Output that is not unique
The terms read here warn that output may repeat:
- Runway: “OUTPUTS MAY NOT BE UNIQUE AND SIMILAR OUTPUTS MAY BE GENERATED FOR OTHER USERS.”
- OpenAI: “output may not be unique and other users may receive similar output from our Services.”
- Google, Gemini API: Google “may generate the same or similar content for others and that we reserve all rights to do so.”
- Google Cloud: a Generative AI Service “may, in some scenarios, produce the same or similar Generated Output for multiple customers.”
For an ad, uniqueness matters most in three places: a generated element that becomes a recurring brand asset (a mascot, a signature location, a pack-shot environment), anything the brand may later want to register as a trademark, and any shot a competitor could plausibly prompt for. Those are the elements to make, or substantially finish, by hand.
A simple decision rule follows. If an element only needs to look right for one flight of one campaign, a generated version may be acceptable once the other questions are answered. If it needs to be recognizably and exclusively the brand’s for years, plan it as human-made work (drawn, built in 3D or filmed) and use AI-assisted steps only to support that work.
Training on your uploads: where client material goes
In the terms read here, individual and free tiers let the vendor use what you upload, and what the tool generates, to improve its models; business tiers do not, unless the customer agrees. Because client material includes unreleased products, scripts and talent images, this question decides whether a tool can touch client work at all.
| Terms | What they say about training on your content |
|---|---|
| Runway Terms of Use | Inputs and Outputs “may be used by the Company to train and improve its AI models,” under a “non-exclusive, irrevocable, perpetual, worldwide, royalty-free, fully paid, transferable, sublicensable” license |
| Runway Enterprise Services Terms | “Runway may not use Customer Content as training data for the Services.” |
| OpenAI Terms of Use | Content may be used to “provide, maintain, develop, and improve our Services”; you can opt out of training by following OpenAI’s instructions |
| OpenAI Services Agreement (section 4.2) | “OpenAI will not use Customer Content to develop or improve the Services, unless Customer explicitly agrees to such use.” |
| Gemini API, Unpaid Services | Google uses submitted content and responses to “provide, improve, and develop” its products; “human reviewers may read, annotate, and process your API input and output”; “Do not submit sensitive, confidential, or personal information to the Unpaid Services.” |
| Gemini API, Paid Services | Google “doesn’t use your prompts (including associated system instructions, cached content, and files such as images, videos, or documents) or responses to improve our products” |
| Google Cloud Service Specific Terms (section 18) | “Google will not use Customer Data to train or fine-tune any AI/ML models without Customer’s prior permission or instruction.” |
What counts as client material
In ad production, the inputs that matter are rarely the text prompt. They are the reference frames, plates and files a team uploads to guide a generation or a cleanup:
- Product photos, CAD renders and packaging files for products that have not launched.
- Storyboards, scripts and animatics that reveal an unannounced campaign.
- Footage and stills of performers, which also raise consent questions.
- Voice recordings used for reference or localization.
- Brand assets such as logos, typefaces and color references.
If any of these goes into a tool whose terms allow training, the material may stay in the vendor’s training pipeline. Runway’s individual terms describe the license as “irrevocable” and “perpetual” and cover inputs “at any point,” so moving to an enterprise plan later may not change the status of material already uploaded. Ask counsel how the terms treat earlier uploads before relying on an upgrade.
Sharing features are a separate grant
Public sharing features can create rights the generation itself does not. OpenAI’s Service Terms say that by sharing an image or video publicly, “such as by sharing it to the Sora feed or uploading it to create a Sora cameo,” you give OpenAI the right to reproduce, distribute, modify, display and perform it to operate and promote the services. Runway’s terms warn that other users may “search for, see, use, modify and reproduce” content you submit to any “public” area of its services. A production’s internal rule can be simple: no client work is posted to a tool’s public feed, gallery or community.
Indemnity: what legal cover exists and where it stops
An indemnity is the vendor’s promise to defend you and pay damages if a third party sues over the tool or its output. Of the terms read here, only OpenAI’s and Google’s business terms, and Adobe Stock for some generated images, extend indemnity to output. The published exclusions catch ordinary ad-production steps, starting with modified output.
| Terms | Indemnity to the user? | Covers generated output? | Key exclusions |
|---|---|---|---|
| Runway Terms of Use | No; the user indemnifies Runway (section 9) | No | Services provided "AS IS" and "WITH ALL FAULTS" |
| Runway Enterprise Services Terms | Yes, for claims that "the Services infringes" a third party's IP | Not stated; Runway makes no warranty on "the legality of Outputs" | Inputs; customer negligence or breach; modification or combination not done or authorized by Runway |
| OpenAI Terms of Use | No; business and organization users indemnify OpenAI | No | Not applicable |
| OpenAI Services Agreement and Service Terms (API, Enterprise) | Yes (section 13.1) | Yes, use or distribution of Output | Knew or should have known; safety features disabled; Output "modified, transformed, or used in combination" with non-OpenAI products; no rights to Input; trademark use in commerce; Beta Services; Prompt-to-Voice |
| Google Cloud Service Specific Terms (section 20) | Yes, for paid Generative AI Indemnified Services | Yes, "unmodified Generated Output" | Knew or should have known; tools disregarded or disabled; use after notice of a claim; trademark use in commerce; no rights to tuning data |
| Adobe Stock license page | Some generated outputs | Text to Image outputs from the Adobe Stock Generate page "may come with IP indemnification" | Scope set in Adobe's Terms and Conditions |
Google’s two indemnities
Google Cloud announced a “two-pronged” approach on October 13, 2023: one indemnity for Google’s use of training data, one for generated output. The current Service Specific Terms keep both. The output indemnity applies to “an unmodified Generated Output from a Generative AI Indemnified Service,” and that defined term covers only services listed on Google’s indemnified services page whose use “is not provided to Customer free of charge.” That page, last modified July 20, 2026, lists the Gemini Enterprise Agent Platform API (formerly Vertex AI API) “used with generally available versions” of models including Gemini, Imagen and Veo.
The training-data indemnity covers allegations that Google’s use of training data to build its pre-trained models infringes, but “does not cover allegations related to a specific Generated Output.” The 2023 announcement also set the customer’s part: the cover applies only if you “didn’t try to intentionally create or use generated output to infringe the rights of others.”
OpenAI’s output indemnity
The Services Agreement indemnifies business customers against claims that “the Services infringe any third-party IP Right” (section 13.1). The Service Terms extend this, for API and Enterprise customers, to claims that “Customer’s use or distribution of Output infringes a third party’s intellectual property right.” Six exclusions follow, and the third is the one that matters in an edit suite: the indemnity does not apply where “Output was modified, transformed, or used in combination with products or services not provided by or on behalf of OpenAI.” Beta Services “are excluded from any indemnification obligations,” and so is Prompt-to-Voice.
Runway’s position
Runway’s individual Terms of Use give no indemnity to users and require users to indemnify Runway. Its Enterprise Services Terms indemnify the customer against claims that “the Services infringes” a third party’s IP rights, excluding claims arising from inputs and from modification or combination “not performed or approved by Runway or specifically set out in the Documentation.” The same terms state that Runway “makes no representations or warranties with respect to the legality of Outputs.” Read together, enterprise cover is about the service itself; it does not promise that a given output is clear to use.
Unmodified output and the edit suite
Almost no generated shot reaches air unmodified. It is graded, cropped to 9:16, composited with a filmed plate, retimed and encoded. Google’s output indemnity names “unmodified Generated Output,” and OpenAI’s excludes output “modified, transformed, or used in combination” with other products. Whether a grade or a reframe counts as modification under a given clause is a question for counsel, and it should be asked before the shot is approved, not after a claim arrives.
What an indemnity pays for, and what it does not
An indemnity pays for a third party’s claim, not for the brand’s own costs. OpenAI’s covers “liabilities, damages and costs (including reasonable attorneys’ fees) payable to a third party”; Runway’s enterprise indemnity covers “damages and reasonable attorneys’ fees finally awarded against Customer.” Neither describes paying for a recut, a replacement shot or media spend lost when a spot is pulled from air.
The process matters too. OpenAI’s Services Agreement (section 13.4) asks for “prompt written notice” of a claim and gives the vendor “sole control of defense and settlement,” with the right to choose counsel. Whoever would receive a claim letter, at the brand or the agency, should know that the vendor has to be told quickly and that the vendor will run the defense.
Who should be the Customer
Business terms run between the vendor and the organization that signs, so the indemnity follows the contract, not the brand on the screen. That leaves three basic arrangements, each with a different result:
- The brand holds the account and gives the agency or studio seats on it. The brand is the Customer, its data-use terms apply, and any indemnity runs to the brand.
- The agency or studio holds the account and uses it for several clients. The vendor’s promises run to the agency or studio, and the brand’s protection depends on what the production contract passes through.
- A freelancer uses a personal account. Individual terms apply, usually with no indemnity to the user and broader data use.
None of these is wrong in itself, but each should be chosen deliberately and written into the production contract before the first generation.
Liability caps
Caps limit what you could recover for most other failures:
- Runway’s Terms of Use cap liability at the greater of what you paid in the twelve months before the event giving rise to liability, $100, or a remedy set by the statute under which a claim arises.
- OpenAI’s Terms of Use use the same structure: the greater of what you paid in the prior 12 months for the service or $100.
- OpenAI’s Services Agreement caps each party’s liability at the total paid in the prior twelve months, but carves out indemnification obligations, and says the service-specific indemnity “is not subject to any liability cap.”
Example (fictional). Suppose a freelance editor generates a background plate for a client spot on a free individual account and pays nothing in the twelve months before a claim. Under either set of individual terms above, the contractual figure is the greater of $0 and $100, which is $100 (Runway’s clause also preserves a statutory remedy where one applies). The same terms give the user no indemnity at all, and Runway’s require the user to indemnify the vendor. If the shot had been made under a paid business agreement with output indemnity, the cover would still depend on whether the grade and composite count as modification.
Likeness, voices and trademarks in the terms
The terms put responsibility for real people and brands on the user. They prohibit reproducing a person’s likeness without consent, exclude trademark claims from indemnity, and leave releases and clearances to the customer.
- Likeness. OpenAI’s Service Terms: “You may not use Visual Capabilities to reproduce the likeness of any person without express consent and all necessary rights.” Runway’s Terms of Use prohibit content that infringes any “right of publicity” and posting a photograph of another person “without that person’s permission.”
- Voices. OpenAI’s Prompt-to-Voice may not be used “to reproduce any person’s voice without express consent and all necessary rights,” customers must disclose that generated speech is AI-generated, and the feature is outside OpenAI’s indemnity.
- Releases. Runway’s Enterprise Services Terms make the customer responsible for having “all rights, licenses, and permissions required to provide Inputs,” “including publicity clearances and releases,” and for “evaluating Outputs for legality,” including “commercially reasonable human review efforts as appropriate.”
- Trademarks. Google’s output indemnity does not apply to allegations “based on a trademark-related right as a result of Customer’s use of such Generated Output in trade or commerce”; OpenAI’s output indemnity has the same kind of exclusion.
For an advertiser, the trademark exclusion is the one to note. An ad is commercial use, so a generated shot that shows a recognizable logo, product design or packaging needs the same clearance review as filmed footage. Performer consent, union terms for digital replicas and state likeness laws are separate questions from the tool’s terms; Blendpipe’s AI practices page sets out the consent rule used in its own plans.
The five-question check before approving an AI step
Run five questions against the exact terms in force for the account that will generate the shot, and stop or escalate if any answer is missing. The check takes the vendor’s own documents as evidence, not a sales page or a comparison article.
- Which plan, and which terms govern it? Identify the account (whose name, which tier, paid or free), then find the governing document and its date. Red flag: a personal, trial or free account used for client work, or business terms the brand has never seen.
- What do the terms say about output ownership and uniqueness? Quote the clause. Note any “if any” or “to the extent permitted” language and the similarity warning. Red flag: the shot is meant to become a recurring brand asset that must be distinctive.
- Can the vendor train on what you upload or generate? Find the data-use clause and any opt-out, and list the client files that would be inputs. Red flag: training is allowed and the inputs include unreleased products, unannounced scripts or performers.
- What does the indemnity cover, and what does it exclude? Note who the “Customer” is, whether output is covered, the modification and trademark exclusions, beta exclusions and any cap. Red flag: the output will be graded, composited or reframed and the cover is limited to unmodified output.
- What are the likeness, voice and trademark rules? Find the prohibitions and who is responsible for releases. Red flag: a real person’s face or voice, or a recognizable mark, with no written consent or clearance.
If an answer cannot be found in the documents, the answer is “unknown,” and an unknown goes to counsel before the step is approved.
Questions to send the vendor in writing
Some answers are not in the public terms, especially for business plans negotiated on an order form. Ask in writing and keep the reply with the terms:
- Which document governs our account today, and has an order form changed any of it?
- Are our uploads and outputs used for training, evaluation or human review, and can that be switched off for the whole organization?
- How long are uploads and outputs stored, and can we delete them?
- Does your indemnity cover output after we grade, crop, composite or re-encode it?
- Which models or features on our plan are excluded from indemnity, such as previews or betas?
- Do you write C2PA Content Credentials or a watermark into outputs, and can either be turned off?
The last question links the terms to disclosure. The IPTC’s FAQ on C2PA (July 2026) says OpenAI, Google, Adobe and Microsoft embed Content Credentials in their AI-generated content, and platforms such as YouTube read that metadata when they decide on AI labels.
Recording the answers in the rights sheet
Write the five answers where the rest of the clearance record lives: the rights sheet for each shot that uses an AI step, and the provenance log for what was done and who reviewed it. A rights sheet lists every licensed asset with its licensor, term, media, territory and renewal date; an AI tool’s terms are one more license on that list.
| Rights sheet field | What to write for an AI step |
|---|---|
| Asset | Shot number, timecode and the AI step (generated plate, cleanup, matte, upscale) |
| Licensor | The vendor and the legal entity named in the terms |
| Governing terms | Document name, version date and plan tier, with a saved copy |
| Grant | Commercial use permitted; ownership language quoted |
| Training | Allowed or not; opt-out status; which client files were inputs |
| Indemnity | Covered or not; who the Customer is; exclusions that apply to this shot |
| People and marks | Likeness, voice or trademarks present; consent or clearance reference |
| Reviewer | The person who checked the terms and the output |
The rights and usage page explains the rest of the sheet, and the same record sits beside licensed stock: stock footage licenses for ads covers that side, where the licensor usually grants rights it holds rather than declining to claim them.
Example (fictional). Orlenna is an invented brand and the Ridge Bottle spot is an illustration, not a client project. The 30-second spot at 24 fps has two AI-assisted steps: a cleanup that removes a power line from shot 1 (0 to 4 s, 96 frames) and AI-assisted rotoscoping for the hand matte in shot 5 (17 to 20 s, 72 frames), finished by a person.
Both steps upload client material: shot 1’s plate and shot 5’s plate of the hand that will hold the product. Question 3 therefore decides the tool. Under the training clauses quoted above, a tool on terms that allow training on inputs fails the check unless the client agrees in writing; a tool on terms that rule out training passes.
Neither step generates a new picture from a prompt, so question 2 matters less: the frames are filmed footage, cleaned and separated. Question 4 still applies, because the cleanup output is modified and composited. Each step gets one rights sheet row with the terms’ date and the reviewer’s name, and the 168 affected frames (96 + 72) are listed in the provenance log.
How often the terms change, and how to keep a record
Terms for AI tools change often, so the record that matters is the version in force on the day the shot was generated. Save a copy, note the date and recheck before each new campaign or renewal.
The documents read for this guide show the pace. Runway’s Terms of Use were last updated October 23, 2026 and its Enterprise Services Terms June 1, 2026. OpenAI’s Service Terms were updated November 13, 2026. Google Cloud’s Service Specific Terms list a long run of earlier versions, the latest modified November 18, 2026, and its indemnified services list was last modified July 20, 2026.
Runway’s terms add a clause worth knowing: if you continue to use the services after a change, that “constitutes your acceptance.” Business terms can be steadier. OpenAI’s Services Agreement says the service-specific indemnity in force on the effective date is part of the agreement, and that OpenAI “may not materially reduce Customer’s protections” under it “without Customer’s written agreement.” Whether a given plan has a clause like that is one more thing to record. In practice:
- Save the terms at generation. A PDF of each governing document, named with the date, goes into the project folder.
- Log the account and tier. The rights sheet records whose account generated the shot.
- Recheck at each reuse. A spot recut for a new flight or market a year later may sit under different terms than the original.
- Watch the indemnified lists. Google’s cover depends on a listed service and a generally available model version; a preview model may not qualify.
- Re-run the five questions when a vendor announces new plans, features or model versions.
What the production contract should say about AI tools
The vendor’s terms bind whoever holds the account; the production contract decides what the brand can rely on. These are the points to put in front of counsel when a brand commissions a spot that may use AI-assisted steps:
- A list of AI steps and tools, by shot, approved before production, with a rule that new tools need written approval.
- A training rule: client material is not put into tools whose terms allow training on it unless the brand agrees in writing.
- Copies of the governing terms, dated, delivered with the rights sheet.
- Account ownership: whose account generates the shots, and whether vendor indemnities are passed through or assigned where the terms allow.
- Likeness and voice consent: no real person’s face, body or voice generated, altered or cloned without written consent, and union terms where they apply.
- Disclosure and provenance: who decides on platform labels, and whether Content Credentials written by the tools are kept.
- Authorship information: a shot-by-shot statement of which parts are human-made, so counsel can decide what to register.
These points mirror the commitments on Blendpipe’s AI practices page, but they belong in any production contract, whoever makes the spot. How ownership of footage, project files and AI elements is usually split between brand and producer is covered in who owns the footage, project files and AI elements.
Common misreadings of AI tool terms
The trap in each case is treating one clause as the whole answer. Each row sets a tempting assumption against what the published terms read here actually say.
| Misreading | What the terms say |
|---|---|
| “The terms say we own it, so it is copyrighted.” | Ownership is contractual and often “if any”; the Copyright Office says purely AI-generated material is not protected. |
| “Commercial use is allowed, so the shot is cleared.” | Permission to use says nothing about third-party rights; likeness, trademark and copyright clearance remain with the user. |
| “We are on an enterprise plan, so the output is covered.” | Runway’s enterprise indemnity covers the service, not the legality of outputs; OpenAI and Google exclude modified output and trademark use. |
| “The indemnity protects the brand.” | It protects the “Customer” who signed; if an agency or studio holds the account, the brand is not the party. |
| “A free trial is fine for a quick test.” | Free and unpaid tiers can allow training and human review of inputs; Google’s Gemini API terms say not to submit confidential information to unpaid services. |
| “Beta features are just early access.” | OpenAI’s Service Terms exclude Beta Services from indemnification. |
| “Once we opt out, everything is private.” | Opt-outs and plan changes apply under the terms’ own wording; earlier uploads may be under a perpetual license. Ask counsel. |
The common thread: the terms describe the vendor’s obligations, not the brand’s clearance. Clearance still runs through the brand’s own legal review, the same way it does for filmed and licensed material.
What to do next
Before the next AI step is approved for a paid campaign:
- Identify the account, tier and governing terms for every AI tool in the plan, and save dated copies.
- Run the five questions and escalate every “unknown” to counsel.
- Keep client files out of any tool whose terms allow training, unless the client agrees in writing.
- Decide who should be the “Customer” on business terms, so the indemnity runs to the right party.
- Record the answers in the rights sheet and the step, reviewer and frames in the provenance log.
Blendpipe’s AI-assisted production page describes where AI-assisted steps are used in its plans and how each is reviewed, and the AI practices page sets the policy on client material and training. More articles on AI steps in production are collected in the AI in production hub. If you are planning a spot and want the AI steps, terms checks and rights sheet written into the production plan from the start, send a brief.
