Blendpipe homeSend a brief
Menu

AI practices: use, review, disclosure and consent

Where AI-assisted tools may be used, who reviews the output, and how it is disclosed and recorded.

Use the AI disclosure checklist
Every AI-assisted step
Logged with shot, tool category and reviewer
Before use
A person reviews the output
Never without written consent
A real person's face, body or voice
Your material
Not used to train AI models
A magnifying glass rests on a glowing light table over a strip of 35mm film negatives, and the one frame under the lens holds a drawn blue sparrow that the neighboring frames do not.

Blendpipe uses AI-assisted tools for specific production steps, lists every one of them, and has a person review the output before it is used. This page is the policy. What the work looks like in practice is on AI-assisted production, and how it appears in a plan is shown in the fictional example campaign.

Our commitments

  1. Every AI-assisted step is listed in the provenance log: the step, the shot it touched, the tool category and the person who reviewed it.
  2. A person reviews AI-assisted output before it is used.
  3. No real person’s face, body or voice is generated, altered or cloned without that person’s written consent, and the union’s terms where they apply.
  4. Client material is not used to train AI models, and it is not put into tools whose terms allow training on it unless the client agrees in writing.
  5. Disclosure follows each platform’s current rule and applicable law, and Content Credentials are kept or added where the tools and the destination support them.
  6. You are told which parts of the work are human-authored, because AI-generated material may not be protected by copyright.

Where AI-assisted tools may be used

Use What it does Who reviews it In the final frame?
Concept frames Early pictures to explore a look in the treatment Director or designer No; labeled AI-assisted in the plan
Previs Rough moving pictures to plan camera and timing Director No
Rotoscoping Mattes that separate a subject from its background Compositor, who finishes the edges As part of a composite
Cleanup Removing wires, signs, power lines or marks Compositor, frame by frame Yes
Upscaling Raising the resolution of an image or plate Compositor or finishing editor Yes
Caption and translation drafts First drafts of captions and subtitles A fluent reviewer Yes, after correction

Each step is written into the plan before it happens, with its disclosure decision.

Where they are not used without written agreement

Some uses change what an audience takes to be real. They are not used unless you agree in writing and, for a real person, unless that person consents in writing:

  • Synthetic people (performers who do not exist) and synthetic voices.
  • Digital replicas of a real performer, or any change to a real person’s face, body or voice.
  • Realistic footage of real places or events that did not happen.

Testimonials from people who do not exist are not made at all. The FTC’s 2024 rule on fake reviews bars testimonials that misrepresent that they are by someone who does not exist, “such as AI-generated fake reviews” (FTC announcement).

Human review

The plan names the reviewer for each step by role, and the provenance log names the person. They check the output at full resolution: artifacts, the accuracy of anything a claim depends on, unintended resemblance to a real person, brand rules, and whether it now needs a disclosure. Output that fails review is redone or replaced with conventional work.

Tools and their terms

Plans name tool categories (generative image model, AI-assisted rotoscoping), not brands. Where a public product is named, it is named with its published terms, never as a partner. For example, Adobe states that its Firefly models are “trained on licensed content from Adobe Stock and public domain content” (Adobe Firefly), and Google DeepMind states that Veo videos are “marked with SynthID” (Veo). Tool terms on commercial use, output ownership and training on uploads are checked before a tool touches a project.

Disclosure by platform and law

Checked November 22, 2026. These rules change, so each plan rechecks them.

Rule What it says
YouTube Creators must disclose realistic AI-generated or altered content, such as a real person appearing to say or do something they did not, or altered footage of real events or places. Not needed for unrealistic content, minor edits such as color or beauty filters, or production help. YouTube may apply an AI label automatically to content with C2PA metadata.
TikTok Creators must label AI-generated or significantly edited content showing realistic-looking scenes or people; unlabeled content may be removed, restricted or labeled. AI content of a public figure supporting products, or of a private figure without consent, is not allowed.
Meta The “Made with AI” label became “AI info” on July 1, 2024, based on industry-standard indicators and self-disclosure; content only edited with AI shows it in the post menu. Since January 11, 2024, social issue, electoral and political ads must disclose photorealistic AI or digitally altered imagery or realistic audio; immaterial uses such as resizing are exempt (Meta political ads policy).
Google Ads Election ads with synthetic or digitally altered content must be disclosed; changes inconsequential to the ad’s claims, such as cropping or color correction, are exempt (Google Ads policy).
New York S8420-A Producers of commercial ads must conspicuously disclose a synthetic performer they have actual knowledge of: $1,000 for a first violation, $5,000 after. Exempt: ads for expressive works, audio ads, and AI used solely for translation. Effective 180 days after signing on December 11, 2025 (June 9, 2026 by our count).

Each production plan proposes a disclosure decision for every step; your counsel makes the final call. The AI disclosure checklist walks through the same questions.

Content Credentials and their limits

Content Credentials, built on the C2PA open standard, attach signed statements to a file about how it was made; AI actions are recorded in a field called digitalSourceType, and a hash binding makes the record tamper-evident. Blendpipe keeps the credentials that tools write and adds them where the pipeline and destination support it. Two limits matter. The C2PA explainer says plainly that provenance metadata can be removed (watermarks or fingerprints help recover it), and credentials “do not provide value judgments” about whether content is true. That is why the provenance log is also delivered as a separate document.

The US Copyright Office’s registration guidance (March 2023) says copyright “can protect only material that is the product of human creativity”, and material generated from solely a prompt must be disclaimed in a registration (guidance). Part 2 of its AI report (January 29, 2025) concludes that prompts “do not alone provide sufficient control”, that AI used “to assist rather than stand in for human creativity does not affect” protection, and that human expression perceptible in the output, creative selection and arrangement, and modifications are protected.

The provenance log shows, shot by shot, which parts people made (filmed footage, drawn animation, built models, the edit) and which steps were AI-assisted. Your counsel can use it to decide what to register and what to disclaim.

When union performers are involved, their contract adds terms. Under the 2025 SAG-AFTRA Commercials Contracts (April 1, 2025 to March 31, 2028), the Joint Policy Committee overview lists 48 hours’ notice before creating or using a digital replica, consent through a Digital Replica Rider or a separate signed writing, and 1.5 times scale session fees when a replica generates a performance. The memorandum of agreement bars using covered performances to train generative AI without the union’s consent, requires deleting a replica within 90 days and certifying it in writing when there is no consent to keep it, and requires a performer’s consent before a generative AI system is prompted with their name or image. These are industry terms; the plan accounts for them whenever union talent is used.

Your material and training

Your footage, product drawings, scripts, recordings and brand assets are not used to train AI models, and they are not put into tools whose terms allow training on uploads unless you agree in writing. Questions about your files go to support@blendpipe.com, and new projects start with a brief.