Deep Voodoo: How Trey Parker and Matt Stone Built Hollywood's Most Ethical AI Studio
.jpg?w=3840&q=90)
Share this post:
Deep Voodoo: How Trey Parker and Matt Stone Built Hollywood's Most Ethical AI Studio
Trey Parker and Matt Stone founded Deep Voodoo in 2020 during the COVID shutdown and built something the AI industry rarely produces: a likeness studio that has refused to take jobs without performer authorization since its first day of operation. In May 2026, the regulatory framework Hollywood is now building around AI finally caught up to what Deep Voodoo has been doing for six years.
What Deep Voodoo Actually Does
Deep Voodoo operates a Venice studio with nine cameras. For each project, the studio captures approximately 300,000 images of the subject and uses that data to build a custom AI model specific to that performer. Building each model takes up to one month per person. CEO Afshin Beyzaee has stated the studio has never scraped public images for likeness data.
The process is the inverse of how most AI likeness work is done. Generic datasets scraped from the internet produce models that can approximate a face without the subject's knowledge or agreement. Deep Voodoo's approach means every model it builds requires a specific session with the actual person, which makes unauthorized use structurally impossible.
The Custom Model Pipeline
What happens between the 300,000-image capture and a finished production model takes up to four weeks. The studio runs the images through preprocessing that normalizes lighting, angle distribution, and facial geometry across the full set. The resulting training data covers the full range of expressions, head positions, and lighting conditions the model will need to handle in production.
The custom model is calibrated to the subject's specific facial structure, which means it will not generalize to other subjects and cannot be misused as a generic deepfake tool. Quality validation involves comparing model outputs against reference footage before the model is approved for production work. That validation step is what makes the one month build time necessary.
The Consent Policy
Deep Voodoo refuses all jobs without proper licensing and performer authorization. "It's very inappropriate to be taking and making use of someone's likeness without their permission", Beyzaee told The Hollywood Reporter. The policy is a business model, not a legal hedge.
The studio received $20 million in funding led by Connect Ventures, the joint partnership between CAA and NEA. That investment, from Hollywood's largest talent agency, signaled industry trust in the consent approach from the company's founding. Matt Stone has described the studio's AI work in practical terms: "It means maybe the show's better".
The Connect Ventures Investment
CAA's participation in Connect Ventures brought more than capital to Deep Voodoo. CAA represents a large share of the talent whose likenesses might be used in production work, which gives Deep Voodoo a direct pipeline to the consent sessions the custom model pipeline requires. An agency relationship with the talent makes scheduling a 300,000-image capture session operationally straightforward in a way that a cold approach to performers would not be.
NEA, the venture firm, brought technology investment experience to complement CAA's entertainment relationships. The joint structure means Deep Voodoo has backers who understand both the technical side of AI model development and the industry relationships that make consent based production work at scale.
What the Studio Has Built
Original: Trademark rights owned by Comedy Central. SVG version: Lommes, Public domain, via Wikimedia Commons
Deep Voodoo's project list covers music, advertising, television, and film. Kendrick Lamar's "The Heart Part 5" music video used Deep Voodoo's face morphing technology to show Lamar transforming between celebrity likenesses. The studio produced the Dunkin' Donuts Super Bowl advertisement and created the Donald Trump deepfake used in the South Park Season 27 opening.
For music, Deep Voodoo restored Billy Joel to a younger appearance in the "Turn the Lights Back On" video. Its oldest credited work is the Bill Clinton scene in "Ted" (2012). The "Sassy Justice" web series, which uses deepfake technology to parody public figures, was built on explicit consent from the subjects or their estates in every case.
The Heart Part 5 in Detail
Kendrick Lamar's video for "The Heart Part 5", released in May 2022, showed Lamar's face morphing into those of Kanye West, Will Smith, Nipsey Hussle, O.J. Simpson, Kobe Bryant, and Jussie Smollett. Each transformation required a separate consent or estate authorization process. The morphing sequence used Deep Voodoo's custom model technology rather than a generic face-swap tool, which produced the quality and detail the video required.
The production challenge was not technical but logistical. Getting authorization for six likenesses before a music video release requires months of coordination. Deep Voodoo's pipeline made that coordination the standard starting point for the project, not an afterthought added after the creative work was complete.
Why 2026 Is the Inflection Point
The Human Provenance in Film standard, launched at the Cannes Film Market on May 12, 2026, codifies what Deep Voodoo has been doing operationally since its founding. The standard requires AI disclosure in distribution and sales chains. Deep Voodoo's pipeline already produces that documentation as a byproduct of its consent process.
The California digital replica law requires performer consent for AI generated likenesses used in commercial productions. Cate Blanchett and RSL Media's consent framework extended that argument to a human-rights dimension. The SAG-AFTRA four year deal built consent requirements into studio contracts. Deep Voodoo built all of this into its workflow before any of it was required.
Photo by Harish Karumanchi on Unsplash
The Human Provenance in Film Standard
The Human Provenance in Film standard was launched at Cannes on May 12, 2026, at the Marché du Film. The standard requires that any use of AI generated or AI enhanced performance in a film be disclosed throughout the distribution chain, from the production company to the sales agent to the exhibitor. The disclosure obligation travels with the film rather than sitting only in production documentation.
Deep Voodoo's consent process generates documentation at every stage: the authorization agreement, the capture session records, the model build logs, and the production approval sign-off. A studio distributing a Deep Voodoo project already has the paper trail the Human Provenance standard requires. For productions that used AI likeness tools without that documentation, compliance requires reconstruction after the fact.
AB 2602 and AB 1836
California AB 2602 requires that any use of a living performer's AI replica in a commercial production be covered by an explicit written consent agreement signed before production begins. The performer must receive that agreement before any work starts, and the agreement must specify what the replica will be used for. AB 1836 covers deceased performers and requires written authorization from the performer's estate.
Deep Voodoo's existing consent process satisfies both statutes. The studio requires performer authorization before scheduling a capture session. Its oldest estate-authorized work, the Bill Clinton scene in "Ted" from 2012, predates both laws by over a decade. The studio's legal exposure under the new California framework is essentially zero because the consent process existed before the laws did.
The Regulatory Moment Deep Voodoo Was Built For
California AB 2602 and AB 1836, both signed into law in 2024, created consent requirements that Deep Voodoo had been meeting voluntarily for four years before either law existed. AB 2602 governs living performers: any commercial production using an AI replica of a living actor must have written consent from that actor before production begins, and the consent agreement must specify the intended use. AB 1836 extends the same requirement to deceased performers, requiring estate authorization.
The timing of Deep Voodoo's founding and the eventual passage of these laws reflects how the consent debate moved through Hollywood between 2020 and 2024. Parker and Stone built a consent first studio during the COVID shutdown. The guild negotiations from 2023 produced contracts that embedded consent requirements across covered productions. The California legislature codified both consent frameworks into law by 2024. Deep Voodoo was ahead of each step.
Afshin Beyzaee's Background
Afshin Beyzaee came to Deep Voodoo from a background in visual effects production technology. Before taking the CEO role, he worked on the technical infrastructure for large scale VFX productions, giving him direct experience with the pipeline management challenges that a consent first AI likeness studio requires. His public statements on Deep Voodoo's business model have consistently framed consent not as a limitation but as what makes the studio's work worth buying.
His position in industry discussions on AI regulation has been consistent since 2020. Deep Voodoo participated in early conversations that shaped the SAG-AFTRA language on synthetic performers, and Beyzaee has represented the studio at several industry panels on AI likeness standards. That engagement put Deep Voodoo inside the policy conversations rather than reacting to their outcomes.
What "Ethical" Means as a Business Strategy
Deep Voodoo's consent first model functions as a competitive differentiator because it addresses the primary legal risk that AI likeness work carries. A studio that scrapes public images and builds a likeness model without consent is exposed to California AB 2602, AB 1836, right of publicity claims, and SAG-AFTRA grievances simultaneously. Deep Voodoo's clients acquire a production with documented consent at every step, which eliminates that exposure entirely.
For brand clients, the consent documentation is particularly valuable. A Super Bowl advertisement that uses a celebrity's AI likeness without documented consent creates reputational risk for the brand even before any legal action is filed. Deep Voodoo's pipeline delivers the consent record alongside the creative output. That combination is what drew CAA and NEA to fund the studio rather than a competitor without the consent framework.
The South Park Production Context
The Trump deepfake in South Park Season 27 was produced at a point when Deep Voodoo had over five years of production experience with the technology. Parker and Stone had been funding the studio's development through their own production company since 2020, which meant the technology arrived in their production in a mature form rather than as an experiment.
South Park's history of using political likenesses for satire predates generative AI entirely. The show has depicted dozens of public figures through animation for three decades. Deep Voodoo's AI likeness technology extended that history into photorealistic territory while maintaining the same creative context. Parker and Stone's position as creators who founded the technology studio rather than clients who hired it means the consent framework and the creative decision-making were never separated.
The Consent Model Is Spreading to Audio
The visual and audio sides of AI likeness are converging on the same standard. Matthew McConaughey, who faced unauthorized voice cloning on multiple occasions, eventually licensed his voice to ElevenLabs under explicit consent terms. ElevenLabs, which has partnerships with major Hollywood talent, operates on the same principle Deep Voodoo established for visual likeness.
Filmmakers and creators who want to work with consent based AI audio tools can use the AI FILMS Studio voice workspace, which partners with ElevenLabs and operates under the same model: licensed, attributed, authorized.
Deep Voodoo as a Template for the Industry
Deep Voodoo did not build its consent framework in response to regulation. It built the framework because Parker and Stone's experience in entertainment had made clear that unauthorized likeness use creates litigation risk, reputational exposure, and industry relationships problems that are far more expensive to resolve than the cost of obtaining consent in the first place. The studio's model is a business risk management strategy that also happens to satisfy every regulatory requirement that followed.
That sequence, building compliance ahead of regulation rather than in response to it, is what makes Deep Voodoo useful as a template. Studios that built AI tools on scraped datasets and unauthorized training are now facing retroactive compliance exposure under AB 2602, AB 1836, and the SAG-AFTRA agreements. Deep Voodoo has no such exposure. The consent pipeline it built in 2020 is fully compatible with the regulatory environment of 2026 because it was built on the principle those regulations eventually codified.
The Pricing Structure of Consent-Based Work
Deep Voodoo charges more than studios that build AI likeness tools on public image datasets. The one month build time per custom model, the nine-camera capture session, and the quality validation process before production use all represent real costs that do not exist in pipelines built on scraped data. The studio's clients are paying for the consent record and the legal protection it provides alongside the creative output.
For projects where likeness authenticity and legal defensibility are both important, the higher cost of Deep Voodoo's process is not a premium. It is the baseline cost of doing the work without exposure. The Kilmer production, the Lamar video, and the Super Bowl advertising work all required likeness accuracy at a level that only a purpose built custom model could deliver. The consent process and the quality outcome are the same pipeline, not separate considerations.
Sources
The Hollywood Reporter | TechCrunch | Variety | Deadline
Continue Reading
Video & LipSync
- Video Generator
- Text to Video
- Image to Video
- Start-End Frame to Video
- Draw to Video
- Motion Control
- Video Enhancer
- Video Upscaler
- Video to Video LipSync
- Audio to Video LipSync
- Image to Video LipSync
- Video FaceSwap
- Seedance 2
- Vidu Q3 Pro
- Gemini Omni
- Google Veo 3.1
- Kling 3.0 Pro
- LTX 2.3
- Happy Horse 1.1
- Kling 3.0 Motion
- ByteDance Upscaler
- InfiniteTalk
- InsightFace
_(53532289710)_(cropped2).jpg?w=3840)