In February 2024, OpenAI shook up the world of artificial intelligence with the unveiling of Sora, an instant video generation tool. The promise: turn anyone into a filmmaker by enabling the creation of realistic one-minute videos from simple prompts.
OpenAI, a pioneer in the field ofartificial intelligence, continues to push back the boundaries of innovation with its latest creation: Sora. Following in the footsteps of startups like Runway and tech giants such as Google and Meta, the generative AI model designed by the creator of ChatGPT could revolutionize video generation.
Sora's amazing abilities
The quality of the videos produced by Sora far exceeds previous generations of artificial intelligence. Examples shared by OpenAI and early testers demonstrate the diversity of possibilities offered by this tool. Style adjustments (photorealistic, animated or black and white), shot changes and even the modification of elements within a video are all within Sora's reach, pushing back the limits of video creativity. Unlike many text-to-video conversion models, Sora can produce videos of up to one minute in length, demonstrating its ability to tell an "extended" story. What's more, Sora's results generally avoid the common pitfalls of AI-generated content, such as objects moving in ways that are physically impossible. According to OpenAI, the Sora model could even "extend" existing video clips by filling in missing details, demonstrating its adaptability and versatility.
How does Sora work?
Sora is based on an evolved version of the DALL-E 3 diffusion model combined with the powerful GPT-4 engine. According to Tim Brooks, one of the project's researchers, "Sora learns 3D coherence and geometry by assimilating huge volumes of data, demonstrating an advanced understanding of cinematic grammar."
At the heart of Sora's capabilities is its deep understanding of language, enabling it to accurately interpret prompts and bring them to life in a visually compelling way. OpenAI points out that Sora not only understands user requests, but also "how these elements manifest themselves in the physical world". This linguistic mastery enables Sora to generate characters that express vibrant emotions and scenes that maintain coherence and realism. The result: characters that not only move naturally, but also display a spectrum of emotions, lending an unprecedented layer of depth and realism to AI-generated content. A level of detail in character representation that opens up new possibilities in storytelling and digital art. "We're teaching AI to understand and simulate the physical world in motion, with the aim of forming models that help people solve problems requiring real-world interaction," reads the Open AI blog.
Sora's limits
While Sora's image quality and simplicity are bluffing, some videos nevertheless present what OpenAI described as "AI strangeness", such as humanoid subjects resembling those in a video game, or instances of spatial confusion and inaccuracies in cause-and-effect relationships. OpenAI acknowledges these limitations, noting that the tool may struggle to accurately simulate complex scenes and understand precise descriptions of events over time.
Sora applications
Sora's potential uses are vast. From 3D environments to automatically-generated music clips, not forgetting the creation of video games, Sora offers a wide range of creative possibilities. However, these advances are not without raising questions about the future of video-related professions and certain cultural industries.
AI to unleash creativity
Indeed, with every innovation, the question of the risks engendered by these artificial intelligence tools arises. OpenAI's new video tool is no exception, raising a number of questions both about the impact on industry professionals and the risks of misinformation . In Les EchosMarie-Anne Ferry-Fall, Managing Director of ADAGP (Société des auteurs dans les arts graphiques et plastiques), warns: "Between mid-2022 and mid-2023, there will have been more images created by artificial intelligence than in 150 years of photography!
So should we really be worried? Will AIs replace humans? According to Florian Douetteau, founder of Dataiku (a company specializing in data processing): " AI is to human intelligence what fast food is to gastronomy ". In other words, there's room for everyone. A view shared by Yann le Cun, director of Meta's artificial intelligence laboratory. The man who won the 2019 Turing Award is optimistic. In his view, humans are not destined to be replaced by AIs, but augmented: " Artificial intelligence is a means of amplifying human intelligence, just as machines are a means of amplifying physical strength, and we shouldn't be afraid of artificial intelligence." Finally, according to Aurélie Jean, a French digital scientist and entrepreneur specializing in algorithms and digital modeling, "artificial intelligence could even liberate creativity". In a column on Radio France on November 07, 2023, she explains, "Artificial intelligence can assist creative people, but it cannot replace human creativity, because the latter involves emotional and practical components that algorithms cannot master. "Based on an Adobe study (2016) which revealed that "nearly 40% of creative people's tasks would not serve their creativity", the researcher then envisages AI as a "liberating tool at the service of creativity". So, freed from the non-value-added tasks that will be entrusted to AIs, they will have more time to devote to their creative talents.
Artificial intelligence is the talk of the town, and tydeo has made it a key part of its strategy for the future.
We talk about it in API News (news from companies and economic players in Brittany and Pays de la Loire). "With AI, the video market is already undergoing certain changes. We really want to turn this corner and integrate AI into our platform and our offers to revolutionize the way video is produced".
Read more on API website
