Sora is poised to disrupt multiple industries, including movies, education, gaming, healthcare, and robotics, with potential benefits for individuals (Liu et al., 2024). According to (Lin 2024), Sora may serve to broaden access to mental healthcare services. In the realm of robotics, Sora’s capability to simulate intricate world dynamics proves crucial for advancements in autonomous driving (Guan et al., 2024), such as scenario generation (Li et al., 2024a). Democratization of art creation is a notable affordance of Sora – ‘Everyone will be a filmmaker’ (Cowen, 2024), which holds promise for broadening participation in artistic endeavours. Researchers speculate that medical professionals can use Sora to create informative videos for patients and the public (Waisberg et al., 2024). Sora can also become a potentially powerful tool for education and libraries, offering opportunities for diverse learning modalities, creativity, and critical thinking (Adetayo et al., 2024).

Sora may introduce ethical quandaries concerning privacy and security, copyright, disinformation, and truth integrity. AI’s enhanced video generation capabilities may ease the creation of adversarial attacks (Hannon et al., 2024). Copyright infringement is a paramount concern brought by Sora – it might have leveraged copyrighted materials such as videos for training (Karaarslan & Aydın, 2024). Other concerns raised by the researchers included misuse of AI-generated content (e.g., deep fake videos) and environmental costs. (Rodríguez 2024) discussed the likely ‘ultimate solipsism’ introduced by the rise of ‘self-movies,’ where protagonists are the creators themselves, leading to implications of quality degradation. Moreover, the duality of Sora was expressed by (Cavalcante 2024), who argues that Sora simultaneously indicates democratizing creation, i.e., anyone can become a visual storyteller regardless of their technical or financial status, and ethical dilemmas such as disinformation that puts the integrity of truth at risk.

Researchers started to investigate public perceptions of Sora before its release. In the work of Mogavi et al. (2024), both people’s envisioned applications of Sora and expressed concerns about this emerging technology were captured by qualitatively analysing Reddit discussions. People predicted that Sora would democratize video marketing and gaming content production, enhance educational content by generating illustrative videos, and substantiate people’s creativity and storytelling abilities. In the meantime, Sora threatens creative and artistic jobs and perpetuates social biases from training datasets. We extend this work by examining discussions on a different online platform, comparing discussions around Sora and Midjourney, and focusing on strategies to regulate governance challenges.

To address our research questions, we manually gathered 292 comments under four popular posts on X discussing public perceptions of different aspects of Sora. The first and second posts contained Sora videos released by OpenAI – the first post was by a prominent tech influencer, while the second post was by OpenAI itself, potentially capturing views from different stakeholders. The third post compared Sora videos and Midjourney pictures created with the same prompts, emphasizing visual similarity. By including discussion about and comparisons with Midjourney, a text-to-image AI tool, we might be able to uncover how text-to-video models like Sora presented unique challenges to AI regulation and society compared to text-to-image models. The fourth post was by an artist complaining about Midjourney’s copying behaviour, highlighting regulatory challenges. The four posts were chosen in a way that represents different stakeholders (OpenAI, tech influencers, artists, and the general public) and compares between different GenAI models (text-to-video vs text-to-image models). These posts selected exhibited high levels of engagement, indicated by the large number of views and comments. Descriptions of the posts are presented in Table 1. The comments were put in a hierarchical structure to reflect interaction dynamics. Our dataset includes 93, 139, 39, and 21 comments, respectively, from these posts. We arrived at this truncated list from the original number of comments by excluding comments that included little information. For example, although there were nearly 10 thousand comments under Post 2 by OpenAI by early March when we collected the data, most of the comments were short and not informative (e.g., ‘This is great’).

We employed a thematic analysis (Braun & Clarke, 2012) to analyse the comments. The thematic analysis involved identifying recurring themes, patterns, and categories within the comments, enabling us to gain insights into the predominant discussion topics. Through qualitative analysis, we aimed to uncover the diversity of opinions, concerns, and perspectives expressed by users regarding Sora (RQ1) and its implications for governance and society (RQ2). Emerging themes included people’s assessment of Sora videos, Sora’s impact on industry and profession, governance challenges, and potential solutions to the challenges. Under governance challenges, subthemes included OpenAI’s for-profit nature, replacing humans, data copyright issues, and fabricated content. We regularly discussed to reach a consensus on the themes and organize the themes and subthemes into a hierarchical structure. Below, we use anonymized quotes to illustrate our findings. We retained grammatical errors and typos in the quotes but partially censored swear words with [***] and removed emojis.

People had varied assessments about the video quality Sora afforded. While some regarded Sora’s video generation capability as revolutionary, others pointed out failures in simulating the physical world and its lack of creativity.

Many people were amazed by Sora’s video generation quality: ‘These videos look so real,’ ‘I cannot believe we are here already.’ Some people particularly praised the physical features captured in videos, ‘the physics etc are just too good’. Sora was seen as a revolutionary technology, ‘It has come to a point where it is hard to imagine what AI will be able to do next’. Some even suggested AI may have had access to alternative realities,

what if the AI has already access to alternative realities and is capturing those footages out there and that’s the reason it looks so realistic, and we cannot see the difference.

To some, the video quality was not as impressive, ‘I don’ think it even considers physics in that sense, like if it were a game engine’. Failures in capturing physics were pointed out, e.g., ‘Dog skins and snow doesn’t follow physics correctly’. Many people found that the walking woman’s legs switched sides several times in a video. Some people thought the Sora videos were merely a product of the huge computational power or datasets. Some people were sceptical and thought OpenAI might have used videos that matched the prompt, ‘How do we know its not just giving out unaltered training data that matches the prompt? a dog in the snow video is not uncommon.’

When comparing Sora and Midjourney, people found they converged to similar output, ‘This just shows that most foundational models are going to converge to a similar output. This should give some AI hype pause’. Proprietary data was seen as a key to training high-performance models, whereas all models currently are trained on similar public data, ‘It’s now down to who has the most robust proprietary data. The issue now is all these models are trained off the same data sets’. With all the AI- generated content online as potential training data, the convergence would only get worse, ‘Especially as the internet becomes filled with “noise” in the form of AI outputs, convergence seems inevitable’.

Many expressed the potential for Sora to disrupt the movie and content creation industries. This quote, ‘Sora is incredible and scary’, demonstrated ambivalence.

People agreed that the movie industry would be directly impacted by video generation AI like Sora, ‘Hollywood is about to implode and go thermonuclear,’ ‘RIP the entertainment industry. In a few months, people will make “Avatar” on their phones’. One person even imagined the collaboration between Sora and Midjourney, ‘Midjourney can generate the thumbnails while Sora the movie’. Sora may democratise movie making, making everyone an artist, which has unpredictable implications. One person raved about this democratization process,

thank you, OpenAI has literally democratized cinema. Now, everyone can be an artist and create their own movies going forward. This will drastically reduce the exorbitant expenses of Hollywood, especially for low-quality content with high costs.

However, the democratization of movie creation may also lead to low-quality content,

democratize: Discard all talent through stealing people’s work then allow anyone who don’t know a thing about the art to create their own slop. Wow I love this new democratized cinema world.

This impact may extend to the rest of the media production industry, content creators, and artists. On the bright side, Sora may enable writers to do new forms of storytelling,

perhaps one day regular people, as authors, will be able to create feature-length films from (only) their writing. It would be a refreshing restart to finally wrestle the art form away from Hollywood and allow real stories to be told again.

Sora may democratize content creation for average people,

awesome tool. Tools like this integrated with no-code will truly democratize what regular people can create and push out online. This stuff is amazing and that exponential growth this year will be insane.

The other side of this coin, however, is that Sora may cause job losses for artists, ‘I don’t think y’all realize how many artists you’re f[***] over right now’; the rise of AI content farms,

i would like to miss it because it threatens every single creative profession. Music creation, art, story writing, screenwriting, photography, and film. Unless you are at the highest production level where protections exist, it’ll be impossible for human creators to get traction,

and the frustration of art students,

AI is making me feel I’m wasting my time studying arts. At this point I fear my future is gonna be affected by this medium, the fact that this fear is getting stronger with each new i see about AI “progressing” sickens me, and I’m sure I’m not the only one.

Some people recognized the unfair game between GenAI and human artists, i.e., GenAI has been stealing artists’ work as its training data, ‘It hasn’t democratised anything. It has stolen people’s data without consent to profit of it. Everyone could already be an artist some are just too lazy to’. When an artist shared her experience of being copied by Midjourney, another artist was frustrated and questioned why AI evangelists could not see this dark side of GenAI, ‘Breaks my heart. I’m an amateur artist. I’m sure I’d be pissed if other people bastardised my professional efforts. I simply can’t see why the AI evangelicals don’t see it’.

Some were optimistic despite negative impacts, thinking Sora would not replace artists and content creators in the short term. Sora was not regarded as creative, ‘It’s never replacing real artists. Machines only do loops so that means that they only repeat what has been made. Only Humans are truly creative’. Sora did not recreate the human touch, ‘Even as machinery comes to the scene, there’s always gonna be a demand for man-made stuff. It can never truly recreate that human touch so integral to the craft’.

Governance challenges discussed included OpenAI’s for-profit nature, the authenticity of content, taking people’s jobs, training data, power consumption, and the copyright of AI-generated content.

People favoured regulation and AI literacy education to hold Sora accountable and limit its negative impact.

Prior to the release and widespread use of a product, and its early adoption, its mention in both media and technical texts helps it get attention and investments. Such attention is a double-edged sword as it also leads to the build-up of expectations to an extent beyond what is possible to deliver, leading to damaged credibility and reputation (Brown, 2003). The word ‘hype’ has been used to represent this cycle, and this most famously includes the Gartner tech hype cycle methodology developed by Gartner Inc., a technology research and consulting firm, in 1995; Gartner uses the hype cycle in its consulting reports to understand the evolution of the hype around emerging technologies. Gartner’s hype cycle has five stages: Innovation Trigger, Peak of Inflated Expectations, Trough of Disillusionment, Slope of Enlightenment, and Plateau of Productivity. At the innovation trigger stage, the technology is at the proof-of-concept while in the second stage, there are very few practical implementations despite expectations peaking. Thereafter, the third trough stage is a sharp decline of interest and many business failures. It is only after this that the technology picks up again more maturely and finally reaches mainstream adoption in the final stage (Steinert & Leifer, 2010). We can infer that early expectations being inflated beyond what could be reasonably delivered is the expected norm of evolution of new technologies. It is further understood that the time needed for the full hype cycle might vary – while ‘normal technologies might take five to eight years, fast-track technologies might only take two to four years for maturity’ (Dedehayir & Steinert, 2016). The bell-shaped hype curve produced is due to human irrationality, mainly the following: attraction to novelty and love for sharing, social contagion, and heuristic attitude in decision-making (Fenn & Raskino, 2008). It is also possible that there may be multiple peaks and troughs of visibility and expectations rather than a single one and this may not conform to a hype cycle at all (Dedehayir & Steinert, 2016; Steinert & Leifer, 2010).

Our work on the public discourse around Sora is a suitable representation of tech hype at the inception stage. The actual product has not been publicly released, and only a teaser has been provided by OpenAI. People were nevertheless excited about this proof-of-concept technology, evidenced by heated discussions on Sora-related topics on social media. They had drawn on past experiences with AI to form perceptions and expectations about Sora. We find a directionality to the expectations around Sora, with most comments taking a clear positive or negative position, and expectations and perceptions were logical/fact-based or emotional, resonating with prior research on tech hype expectations (Shi & Herniman, 2023).

Our findings indicate a sizable conflict on similar issues and features between people. A fundamental one is that people had conflicting assessments of Sora’s creativity and video-generation quality: while some thought Sora was a revolutionary technique for capturing physical characteristics, others identified imperfections in physical simulation and noticed similarities between Sora-generated videos and Midjourney-generated pictures. These conflicts were also present in people’s views on the desirability of the technology, its impact on employment, AI law enforcement, its impact on content democratization, the extent of AI regulation, and trust in OpenAI.

Sora could also be viewed as a continuation of previous successful generative AI models. An interesting parallel can be found with virtual influencers, which are artificially generated entities acting as social media influencers, often using hyper-realistic AI-generated imagery, but still human-curated: many of the thoughts evoked in this study included blurring of lines between reality and fiction, democratization using AI, and competition between artificially generated content and human-generated content, resonating with quotes of long-term followers of virtual influencers (Choudhry et al., 2022). This is indicative of something more meta: the public sentiment, perception, and expectations of AI technology may be influenced less by the specifics of that particular product and more by the general zeitgeist of AI. The conversations about Sora were less about the features of Sora itself but more about its potential impact on society, jobs, and freedom, implicating that such fundamental debates on AI have existed in the public sphere (Calhoun, 1993) for some time, suggesting that such public opinion could potentially be harnessed in creating fair legislation as has been envisaged previously (Leane, 2010), and resonating with a more flexible legislative approach (Wiener, 2004).