OpenAI’s Sora app went dark on April 26, 2026, with the Sora API set to discontinue on September 24, 2026 — ending an 18-month period during which Sora was the default AI video reference for film, commercial, and audiovisual production teams. The shutdown didn’t end AI video generation; it reshuffled which tools producers can realistically use, and it intensified the underlying copyright disputes that were already keeping streamer business affairs teams up at night.
This guide walks audiovisual producers — whether you’re cutting a commercial in CDMX or a series for a US streamer — through where AI video production stands after Sora 2, how the surviving toolkit (Veo 3, Runway, Kling, Pika) compares, and how Mexico’s 2026 LFDA framework treats the output. The legal architecture matters because in 2026, AI-derived content has a much harder path to chain-of-title registrability than producers assume.
What Happened to Sora 2 and Why It Matters
OpenAI’s Sora 2 launched in late 2024 as the most capable text-to-video model available to general audiences. By early 2026 it had become a workflow default for previs, storyboarding, and even final-cut shots in commercial and short-form production. The April 2026 app shutdown was driven by a combination of operating costs, mounting copyright complaints, and the expiration of OpenAI’s licensing partnership with Disney that had previously allowed branded characters inside the tool (CineD analysis).
For audiovisual producers who had built workflows around Sora 2, the practical consequence is sharper than it sounds: any project still in post that used Sora-generated assets now sits inside a pipeline whose source vendor has exited the market, with chain-of-title questions left open. That doesn’t make the output unusable, but it does require documentation work most teams skipped during the boom.
The Surviving AI Video Generation Toolkit
Four tools cover most of the post-Sora workflow gap. Each has its own copyright and training-data profile:
Google Veo 3
Veo 3 is the most direct Sora 2 replacement on quality, and the model that has captured most production migrations through Q2 2026. Veo 3 was trained on YouTube videos — openly disclosed by Google and contentious with creators whose content was used without compensation (AI Video Generation 2026 review). For producers, Veo 3’s output is high quality and its IP terms are commercially viable; the underlying training-data dispute remains a chain-of-title question that distributors are starting to ask about.
Runway
Runway has been the longest-running purpose-built tool for AI video in production. The output is shorter-form and somewhat less photoreal than Veo 3, but Runway has invested more visibly in indemnification and licensed-data positioning, which matters when E&O underwriters review the pipeline.
Kling, Pika, and the Open-Source Tier
Chinese model Kling has rapidly improved through 2026 and is now competitive on certain shot types. Pika and open-source models cover specific niches. For commercial and short-form production, the open-source tier offers cost advantages; for streamer-acquired content, the indemnification and training-data uncertainty around these tools makes most distributors hesitate.
The Underlying Copyright Question Hasn’t Gone Away
Sora 2’s shutdown didn’t resolve any of the copyright disputes driving it. Lawsuits over training-data use of copyrighted material continue, and US courts have not converged on a clear answer about whether AI-generated content is itself copyrightable. The Harvard Journal of Sports and Entertainment Law overview captures the current state of the debate.
For audiovisual producers, the practical implication is unchanged: any AI-derived asset entering a production pipeline carries dual copyright risk — the risk that the underlying training data contains unlicensed material the production becomes liable for, and the risk that the output itself is unprotectable. Both risks transfer to the producer, not the AI tool vendor, in most current vendor agreements.
Mexico’s LFDA Framework for AI Audiovisual Content
For productions structured through Mexico — or destined for Mexican distribution — the 2026 Federal Copyright Law reform creates a clearer legal framework than the US has, in some ways stricter and in some ways more producer-friendly. The framework rests on three pillars audiovisual producers must understand:
Article 12 LFDA Authorship Test
Article 12 requires a natural person with identifiable creative contribution for a work to be protected. The Mexican Supreme Court has applied this strictly: autonomously AI-generated material falls into the public domain. For Veo 3 / Runway / Kling output entering a Mexican production, the practical question is whether a human creatively directed each shot — not whether AI was used as a tool. Documentation matters. Full mechanics: AI-Generated Content in Mexican Film: A 2026 US Producer Guide.
The Spanish and Indigenous Dubbing Prohibition
The 2026 Federal Audiovisual Law specifically prohibits AI dubbing of foreign films into Spanish or Indigenous languages. For audiovisual producers, the bright line is that AI vocal generation (Sora’s built-in voiceover, Veo 3’s audio tracks, or specialized voice-cloning tools) cannot be used for Spanish-language dubbing of foreign content intended for the Mexican market. Detailed coverage: Mexico Federal Film Law 2026.
Performer Image, Voice, and AI Replica Rights
The LFDA reform created explicit, separately-consented performer rights over voice, image, and AI-replica use — Articles 87 and 121. Any AI video output incorporating recognizable performer likeness needs separate, scope-defined consent. Synthetic talent generated by the AI tool from text prompts is a different question, but as the output increasingly resembles real public figures, the LFDA reform’s right-of-publicity rules apply. Full talent-side mechanics: Hiring Talent in Mexico: An Image Rights and AI Guide.
Practical Workflow for Audiovisual Producers
A producer’s workflow for AI video in 2026 should look different from the workflow that worked in 2025. Four practical steps:
- Map AI use per deliverable. Track which shots, scenes, audio, and assets used which AI tool, what prompts, and what the human creative contribution was. This is the chain-of-title spine.
- Use tools with the cleanest training-data position when the output will sit in the final cut. For storyboards and previs the training-data question is lower-stakes; for distributed assets it’s central.
- Pre-clear AI use with E&O carriers. Verisk’s 2026 generative-AI exclusion forms have reshaped E&O underwriting — see E&O Insurance for Mexico Film Productions. Carrier conversations should happen before binding, not at claim.
- Route Mexican productions through the LFDA framework. Document human creative direction, plan for the AI-dubbing prohibition, and build separately-consented AI replica clauses into every performer agreement.
What This Means for US Producers Shooting in Mexico
US-led productions shooting in Mexico face a layered compliance question on AI video that productions shooting purely in the US do not. The good news is that the Mexican framework is more developed: Article 12, the dubbing prohibition, and the performer-rights reforms give producers a clearer roadmap than the unsettled US case law. The cost is structural — AI-derived assets entering a Mexico-shot pipeline need to be documented to a higher standard than US producers are used to, and that documentation needs to be in place during production, not assembled at distribution.
The same compliance posture that protects EFICA tax-credit eligibility (Mexico Film Tax Incentives 2026) and survives the inspection regime (Filming in Mexico Inspections) is what streamers, E&O carriers, and Mexican distributors want to see on AI-derived assets. Use the Mexico Production Legal Checklist Generator to map the AI-specific requirements against your project type and structure.
Conclusion: The Tool Changed, the Framework Didn’t
Sora 2’s shutdown is the headline, but it’s not the structural story. The structural story is that AI video generation is now a permanent part of audiovisual production, the copyright framework around it remains contested in the US and is more developed in Mexico, and every streamer business affairs team and E&O carrier is increasingly asking the documentation questions producers used to skip. Building a Mexico-shot production with AI-derived assets in the pipeline? Have ANFEPA structure the AI documentation and contract architecture before post begins.
FAQ: AI Video Production After Sora 2
What happens to projects that used Sora 2 assets after the April 2026 shutdown?
The output remains usable; the chain-of-title and indemnification questions tied to OpenAI’s vendor agreements remain open. Productions in active distribution should document the Sora use case and confirm E&O coverage for the affected deliverables.
Is Google Veo 3 safe to use for commercial production?
Commercially usable, yes. The underlying training-data dispute (Veo 3 was trained on YouTube content) is a chain-of-title question distributors are starting to ask about, but Veo 3’s commercial terms are workable for most production use cases.
Can AI-generated video be copyrighted in Mexico?
No, when it lacks identifiable human creative contribution. The Mexican Supreme Court ruled in 2026 that autonomously AI-generated works fall into the public domain under Article 12 LFDA. AI used as a tool by a human author can still produce protectable work.
Can AI dubbing be used for Spanish-language versions of foreign content in Mexico?
No. The 2026 Federal Audiovisual Law prohibits AI dubbing into Spanish or Indigenous languages. Productions must use human performers for Spanish-language dubbing.
How should audiovisual producers document AI use during production?
Maintain a per-deliverable log of AI tools used, scope of use, prompts, and human creative contribution. Pair the log with vendor-side AI disclosure, human-authorship warranty, training-data representation, and indemnity clauses.
