OpenAI Scraps Ambitious Video Generation Project Sora

Sam Altman made an appearance in Los Angeles for Vanity Fair’s Oscar afterparty earlier this month, while OpenAI was gearing up to license its Sora video generation tool to a Hollywood studio. However, the project was unexpectedly abandoned just weeks later. Reports indicate that Sora encountered significant issues.

A recent article sheds light on OpenAI’s sudden choice to discontinue its highly anticipated video generation product, which was intended to follow in the successful footsteps of ChatGPT. This decision took several major partners, including Disney, by surprise, especially since they had already engaged with Vision, enabling users to craft videos featuring beloved characters like Mickey Mouse and Darth Vader.

Interestingly, Disney executives were informed of this closure less than an hour prior to its public announcement, leaving them shocked by the abrupt change. What they were unaware of was that Sora had secretly become a burden for OpenAI in the months that followed its launch. This was particularly critical as the company began focusing more intently on its operations in light of an impending initial public offering.

During this time, OpenAI was just weeks away from completing a new AI model, known as Spud. They found themselves needing to divert computing resources back to coding and the enterprise products connected to it. AI chips have become incredibly valuable for leading research institutions, and Sora was consuming a disproportionate amount of these resources. The project was not yielding profits, and the demands of users generating content were draining resources that OpenAI required elsewhere.

Sora thus emerged as a costly strategic error, largely due to a key employee involved in the competitive AI talent landscape of Silicon Valley. Altman deemed this decision a necessary sacrifice for the broader aspirations of the company, expressing in a memo to staff that he appreciated their readiness to make “difficult trade-offs” for the sake of the organization.

Originally, the project was close to Altman’s vision of transforming OpenAI into a forefront creator in the AI landscape and serving as a potentially lucrative revenue source. OpenAI first showcased Sora two years ago, presenting fantastical landscapes reminiscent of Hayao Miyazaki’s creations and Salvador Dali’s surrealism. When the standalone Sora app was launched last September, Altman likened it to the groundbreaking release of ChatGPT.

Unfortunately, the app did not achieve the success its creators had hoped for, with users feeling it produced “more AI slop than AI magic.” By year-end, as usage plateaued and financial pressures tightened prior to its IPO, OpenAI executives began reviewing Sora more critically, and their findings were unflattering.

The research team was on the verge of training a new model to enhance video generation within ChatGPT. Unlike language models, which learn from text, video models must interpret an entire moving world, significantly increasing development costs. After assessing the expenses involved, OpenAI opted to cease this initiative.

Moving forward, the company intends to pivot its resources toward new “super apps” that will utilize agent AI tools capable of performing tasks autonomously, such as software creation, data analysis, and travel booking. These productivity-enhancing products have gained popularity among employees, and OpenAI finds itself needing to catch up with rival Anthropic, which poses a threat to its standing in the AI competition. Altman informed his team that the members of the Sora team would shift focus toward long-term projects in robotics.

An OpenAI spokesperson remarked on the company’s commitment to prioritizing computing resources based on long-term economic value, stressing that a disciplined focus would facilitate growth and innovation.

Research teams had the means to track resource allocations internally and were caught off guard by the extent of resources being expended on Sora, especially given its meager revenue and negligible benefits to language model functionality. Secretive operations around the project led former staff to liken it to a startup within a startup.

As 2024 progressed, OpenAI faced challenges in keeping up with competitors. Google’s Gemini gained traction with consumers, while Anthropic’s Claude Code coding tool attracted skilled software engineers due to its ability to write programs with minimal oversight. OpenAI is now rushing to unveil its updated coding product, Codex.

Despite Altman’s aspirations for OpenAI to influence popular culture and entertainment, he engaged former Twitter CEO Parag Agrawal for informal consultations on a social media venture akin to X. Additionally, Altman collaborated with then-Disney CEO Iger to create a partnership that would allow fans to bring cherished characters to life through Sora.

In a preview shown to employees late last September, reactions varied within the company. Some staff were apprehensive about a social media app focused on engagement potentially tarnishing the brand, while others expressed concerns over the safety of user-generated AI videos, even with safety measures in place.

Sora achieved a rapid rise to the top of the App Store after launching, despite being invite-only. Users found it remarkable that they could input any request and receive a generated 10-second video within minutes. The face-upload feature allowed users to star in short, often ridiculous films, and Altman himself seemed indifferent as he volunteered his likeness for sometimes violent or disturbing content.

However, loose copyright controls led to controversial content, such as a video featuring Martin Luther King Jr. that aimed to share dreams on various topics. Following complaints from King’s estate, they demanded the removal of his likeness from the platform.

After initially reaching around 1 million global users, Sora saw its numbers dwindle to under 500,000 in subsequent months, resulting in daily losses of approximately $1 million.

OpenAI made attempts to salvage Sora, announcing in December a multi-year deal with Disney that involved licensing over 200 characters. Disney became a major client of OpenAI, agreeing to invest $1 billion in the startup. Iger expressed that the partnership presented a chance for Disney to engage with the rapid advancements in AI and new media forms. Altman voiced hope that this collaboration would lead to new creative opportunities for users.

The deal with Disney illustrated a practical approach to monetizing intellectual property for AI applications. Interestingly, the day before announcing their partnership with OpenAI, Disney sent a cease-and-desist letter to Google for alleged copyright violations.

In February, Iger revealed during an earnings call that short videos created by Sora would soon be featured in Disney+’s upcoming vertical video section. Disney was also in discussions for a company-wide rollout of ChatGPT.

Recently, OpenAI began testing an enterprise version of Sora with built-in protections for companies like Disney, launching this spring to allow select Disney executives to utilize Sora for marketing and special effects without exposing their work to OpenAI.

However, plans to eliminate Sora had already been in motion. OpenAI fell further behind Anthropic, which made strides that raised concerns about replacing traditional software and briefly impacted market confidence. Recognizing the need to prioritize productivity-driven tools, OpenAI started identifying areas to deprioritize and concluded not to move forward with Sora.

As for Disney’s $1 billion investment, it never materialized, and the collaboration is essentially dormant now. Under new CEO Josh D’Amaro, Disney is actively exploring AI initiatives with various partners.

In response to OpenAI’s exit from the video generation sector, Disney released a statement expressing respect for the decision and appreciation for the collaboration between their teams.