Sora has unexpectedly become accessible to the public.
In the early hours of today, OpenAI’s video generation tool Sora’s internal testing version was leaked. The incident was triggered by some early testers (artists) who became dissatisfied with OpenAI.
Artists used the “most extreme” method to create a brief moment of universal access.
According to leaked open letters, OpenAI had invited hundreds of artists to participate in bug testing, feedback, and experimentation. Despite these artists contributing numerous works without compensation, only a small number of their Sora creations were selected for exhibition through competitions.
Moreover, compared to the enormous PR and marketing value OpenAI gained, the compensation provided to artists was minimal.
The artists felt trapped in an “art-washing” campaign, where each work needed approval from the OpenAI team before sharing, ignoring their demands for creative expression and commentary.
The artists wanted creative freedom, but OpenAI offered only conditional exhibition opportunities, essentially turning them into endorsement tools for Sora.
In anger, some artists leaked Sora’s API on Hugging Face, the world’s largest open-source community platform, making the tool briefly available to the public for several hours.
The artists wrote in their open letter:
We are releasing this tool to give everyone a chance to experience what about 300 artists had: unlimited free access.
We are not against using AI technology as an artistic tool (if we were, we wouldn’t have been invited to participate in the program).
However, we disagree with how this artist program was implemented, lacking artist-friendly support and steering tool development toward corporate interests before potential public release.
By sharing this information with the world, we hope OpenAI will move beyond PR gimmicks and become more open.
According to leaked information, Sora supports video generation in three resolutions: 1080p, 720p, and 360p, with durations of 5-10 seconds.
The model’s codename is Turbo. Previously, The Information reported that Sora has multiple model versions, so there might be other higher/lower-tier versions when officially released.
After the incident, OpenAI quickly took action, shutting down all invitation access and implementing a mute measure on their Discord channel.
Although the access was only available for a few hours, users reportedly generated numerous videos using Sora. Let’s look at some of these works.
Remember the classic Tokyo street demo?
In anime-style rendering, Sora continues to show strong capabilities.
Against a snowy background, a cute bear guards its Coca-Cola.
Notice OpenAI’s unique watermark in the bottom right corner
From Tokyo to Paris, the scene shows a person in a yellow raincoat standing on a busy street.
The classic noodle-eating scene testing wasn’t left out.
A dog leaping in the snow shows impressive accuracy in capturing dynamic moments and animal behavior.
Cat chasing mouse – this time, it’s not Tom and Jerry.
Two faces of robots appear – threatening humanity one moment, helpless the next.
Coincidentally, two months ago, The Information reported that OpenAI was training a new version of Sora, aiming to generate higher quality, longer video clips.
For this, they needed to collect millions of hours of high-resolution video data covering various styles and themes. Earlier, Sora was reported to perform below expectations:
- Slow video generation, initially taking over 10 minutes to generate about 1 minute of footage
- Difficult to use, with filmmakers needing to generate hundreds of clips to find usable ones
- Inconsistent style, objects and characters difficult to maintain consistency across clips
- Physical and anatomical errors present
Based on currently leaked information, while the current version of Sora shows impressive performance, there’s still room for improvement in aspects like video quality.
Moreover, Sora’s generated videos remain “silent movies,” a common criticism, seemingly far from the hyped “reality no longer exists” status. We look forward to more surprises from the upcoming official release.
Just now, according to Techcrunch, OpenAI has officially responded to this leak incident.
An OpenAI spokesperson indicated that Sora is still in testing, and the company is working to find a balance between creativity and safety, preparing for future widespread application.
The spokesperson mentioned that hundreds of artists participating in Alpha testing made important contributions to Sora’s development, with their feedback helping prioritize new features and safety measures.
They specifically pointed out that these artists’ participation was entirely voluntary, with no mandatory feedback or usage requirements.
OpenAI stated they were happy to provide free access to these artists and will continue supporting them through grants, events, and other programs. The company firmly believes AI can be an important creative tool and is committed to making Sora both useful and safe.
Additionally, the spokesperson emphasized that artists need to follow “responsible use” principles when using Sora and strictly maintain confidentiality during development.
However, OpenAI has not clearly defined what constitutes “responsible use” or what information is considered confidential.
Ironically, OpenAI artists had to resort to leaking information to advocate for openness, fighting closed systems with disclosure.
Looking back at this incident, given OpenAI’s past performance, it’s even difficult to judge whether this so-called “artist leak incident” was a pre-planned PR marketing strategy.
If this was indeed an unexpected leak, the massive exposure gained might make OpenAI reconsider their release date.
Since it’s already exposed, why not leverage the momentum and launch Sora earlier?
We’re watching to see if Sora will officially debut this week.
Attached Open Letter
We received access to Sora and were promised participation as early testers, red team members, and creative partners.
However, we believe we were actually being lured into participating in so-called “art-washing” to declare to the world that Sora is a useful tool for artists.
Artists are not your free R&D resources!
We are not your: free bug testers, PR puppets, training data, or validation tokens.
Hundreds of artists provided unpaid labor through bug testing, feedback, and experimental works for a company valued at $150 billion.
Of these unpaid contributions, only a few artists were selected through competition to showcase films created with Sora. These artists’ compensation was negligible compared to the massive PR and marketing value OpenAI gained.
Reject normalizing artist exploitation!
Don’t allow billion-dollar brands to exploit artists for free R&D and PR!
Moreover, every output needed approval from the OpenAI team before sharing. This early access program seemed more about PR and advertising than supporting creative expression and criticism.
We are releasing this tool to give everyone a chance to experience what about 300 artists had: unlimited free access.
We are not against using AI technology as an artistic tool (if we were, we wouldn’t have been invited to participate in the program).
However, we disagree with how this artist program was implemented, lacking artist-friendly support and steering tool development toward corporate interests before potential public release.
We hope that by sharing this information with the world, OpenAI will move beyond PR gimmicks and become more open.
We call on artists to use non-proprietary tools:
Open-source video generation tools allow artists to experiment at the cutting edge without serving corporate interests or being corporate PR tools. We also invite artists to train their own models with their own datasets.
Some available open-source video tools include:
- CogVideoX
- Mochi 1
- LTx Video
- Pyramid Flow
However, we also recognize that not everyone has such resources. Therefore, we call for fair compensation for artists’ expression.
Happy creating,
Some Artists from the Sora Alpha Project