OpenAI shelved its Sora video generation product entirely and folded the team behind it into the company's secretive robotics unit, signaling a major shift in how AI companies are applying video models to the physical world. The move reflects a broader industry trend where companies are betting that AI video models can develop an understanding of real-world physics, potentially unlocking new revenue streams from expensive-to-develop technology.

Why Are AI Video Companies Moving Into Robotics?

The pivot from video generation to robotics isn't unique to OpenAI. Other AI video startups, including Runway, have also made significant pushes into powering robotics in recent years. The reasoning is straightforward: models trained to understand how objects move and interact in video can translate that knowledge into controlling physical systems. This application could represent a lucrative way for companies to generate revenue from models that cost hundreds of millions of dollars to develop.

Black Forest Labs, a German startup known for releasing popular open-weight AI image and video generators, recently announced it was releasing a new model that could also power robots. The company aims to put out an open version of Flux 3 in the coming weeks, which developers can use to generate images, videos, and audio, as well as predict robot actions. CEO Robin Rombach explained the strategic thinking behind this expansion.

"We clearly are stepping from a pure image generation lab to a frontier multimodal AI Lab, which to me is highly important. We are showcasing an intuition which our research team has always had, which is that these models are very general and develop a solid understanding of the real world," said Robin Rombach, CEO of Black Forest Labs.

Robin Rombach, CEO at Black Forest Labs

How Are Video Models Being Applied to Real-World Robotics?

• Vision-Language-Action Models: Most robot foundation models today combine video inputs with written instructions and output motor commands. Google's Gemini Robotics is the best-known example of this approach, which has become the industry standard for translating visual understanding into physical action.
• Direct Video-to-Action Translation: Black Forest Labs' Flux 3 skips the language layer entirely, translating video directly into actions. Rombach argues this approach is more robust because it eliminates an intermediate step that could introduce errors or latency.
• Manufacturing Applications: Black Forest Labs is working with startup Mimic to build a custom version of Flux 3 that powers car-building robots in Audi's factories. The model is particularly effective at controlling robot hands to manipulate flexible parts like window seals and cables onto vehicles.

Stephan Gravert, chief product officer of Mimic, noted that the company is currently testing and deploying the Flux-Mimic AI model with Audi. His team has found that it excels at tasks requiring precision and flexibility. Mimic is planning a full rollout of the technology with Audi by the end of the year, suggesting that AI video models are moving from research projects to production manufacturing environments.

What Does This Mean for the Future of AI Video Technology?

The shift from consumer-facing video generation to industrial robotics represents a fundamental recalibration of how AI companies view their video models. Rather than competing primarily on video quality for content creators, companies like OpenAI are recognizing that the underlying technology has broader applications in automation and manufacturing. This pivot could explain why OpenAI decided to discontinue Sora as a standalone product; the technology may be more valuable as a component of a larger robotics strategy than as a direct competitor to other video generation tools.

Black Forest Labs has raised significant resources to support this expansion. The startup raised $300 million at a $3.25 billion valuation last year, but Rombach indicated the company used more computing power than ever to train Flux 3 and now has over 100 employees. The capital requirements for developing multimodal AI models that can handle images, video, audio, and robotic control are substantial, and the company will likely need additional funding to keep pace with larger competitors.

The robotics pivot also reflects a pragmatic response to market dynamics. Video generation tools face intense competition from multiple startups and established players, making it difficult to achieve sustainable profitability. Robotics applications, by contrast, offer more specialized use cases and higher-value contracts with industrial partners. For companies like OpenAI, folding the Sora team into robotics may represent a strategic decision to focus resources on applications with clearer paths to revenue and competitive advantage.

As AI video models continue to improve, expect more companies to explore similar pivots. The technology that powers realistic video generation is fundamentally the same technology that can help robots understand and interact with the physical world. This convergence suggests that the future of AI video may not be primarily about entertainment or content creation, but about automating complex physical tasks in manufacturing, logistics, and other industrial sectors.