cite from https://openai.com/research/video-generation-models-as-world-simulators

Creating video from text

Sora is an AI model that can create realistic and imaginative scenes from text instructions.

 We’re teaching AI to understand and simulate the physical world in motion, with the goal of training models that help people solve problems that require real-world interaction.

 Introducing Sora, our text-to-video model. Sora can generate videos up to a minute long while maintaining visual quality and adherence to the user’s prompt.

 Today, Sora is becoming available to red teamers to assess critical areas for harms or risks. We are also granting access to a number of visual artists, designers, and filmmakers to gain feedback on how to advance the model to be most helpful for creative professionals.

We’re sharing our research progress early to start working with and getting feedback from people outside of OpenAI and to give the public a sense of what AI capabilities are on the horizon.

Sora is able to generate complex scenes with multiple characters, specific types of motion, and accurate details of the subject and background.

The model understands not only what the user has asked for in the prompt, but also how those things exist in the physical world.

The model has a deep understanding of language, enabling it to accurately interpret prompts and generate compelling characters that express vibrant emotions.

Sora can also create multiple shots within a single generated video that accurately persist characters and visual style.

The current model has weaknesses. It may struggle with accurately simulating the physics of a complex scene, and may not understand specific instances of cause and effect.

For example, a person might take a bite out of a cookie, but afterward, the cookie may not have a bite mark.

The model may also confuse spatial details of a prompt, for example, mixing up left and right, and may struggle with precise descriptions of events that take place over time, like following a specific camera trajectory.

Safety.

We’ll be taking several important safety steps ahead of making Sora available in OpenAI’s products. We are working with red teamers — domain experts in areas like misinformation, hateful content, and bias — who will be adversarially testing the model. We’re also building tools to help detect misleading content such as a detection classifier that can tell when a video was generated by Sora. We plan to include C2PA metadata in the future if we deploy the model in an OpenAI product. In addition to us developing new techniques to prepare for deployment, we’re leveraging the existing safety methods that we built for our products that use DALL·E 3, which are applicable to Sora as well.

For example, once in an OpenAI product, our text classifier will check and reject text input prompts that are in violation of our usage policies, like those that request extreme violence, sexual content, hateful imagery, celebrity likeness, or the IP of others. We’ve also developed robust image classifiers that are used to review the frames of every video generated to help ensure that it adheres to our usage policies, before it’s shown to the user.

We’ll be engaging policymakers, educators and artists around the world to understand their concerns and to identify positive use cases for this new technology. Despite extensive research and testing, we cannot predict all of the beneficial ways people will use our technology, nor all the ways people will abuse it. That’s why we believe that learning from real-world use is a critical component of creating and releasing increasingly safe AI systems over time.

Research techniques.

Sora is a diffusion model, which generates a video by starting off with one that looks like static noise and gradually transforms it by removing the noise over many steps.

Sora is capable of generating entire videos all at once or extending generated videos to make them longer. By giving the model foresight of many frames at a time, we’ve solved a challenging problem of making sure a subject stays the same even when it goes out of view temporarily.

Similar to GPT models, Sora uses a transformer architecture, unlocking superior scaling performance.

We represent videos and images as collections of smaller units of data called patches, each of which is akin to a token in GPT. By unifying how we represent data, we can train diffusion transformers on a wider range of visual data than was possible before, spanning different durations, resolutions and aspect ratios.

Sora builds on past research in DALL·E and GPT models. It uses the recaptioning technique from DALL·E 3, which involves generating highly descriptive captions for the visual training data. As a result, the model is able to follow the user’s text instructions in the generated video more faithfully.

In addition to being able to generate a video solely from text instructions, the model is able to take an existing still image and generate a video from it, animating the image’s contents with accuracy and attention to small detail. The model can also take an existing video and extend it or fill in missing frames. Learn more in our technical report.

Sora serves as a foundation for models that can understand and simulate the real world, a capability we believe will be an important milestone for achieving AGI. 

Technical Report -- Video generation models as world simulators

We explore large-scale training of generative models on video data. Specifically, we train text-conditional diffusion models jointly on videos and images of variable durations, resolutions and aspect ratios. We leverage a transformer architecture that operates on spacetime patches of video and image latent codes. Our largest model, Sora, is capable of generating a minute of high fidelity video. Our results suggest that scaling video generation models is a promising path towards building general purpose simulators of the physical world.

This technical report focuses on

(1) our method for turning visual data of all types into a unified representation that enables large-scale training of generative models,

and (2) qualitative evaluation of Sora’s capabilities and limitations. Model and implementation details are not included in this report.

Much prior work has studied generative modeling of video data using a variety of methods, including recurrent networks,1,2,3 generative adversarial networks,4,5,6,7 autoregressive transformers,8,9 and diffusion models.10,11,12 These works often focus on a narrow category of visual data, on shorter videos, or on videos of a fixed size. Sora is a generalist model of visual data—it can generate videos and images spanning diverse durations, aspect ratios and resolutions, up to a full minute of high definition video.

Turning visual data into patches

We take inspiration from large language models which acquire generalist capabilities by training on internet-scale data.13,14 The success of the LLM paradigm is enabled in part by the use of tokens that elegantly unify diverse modalities of text—code, math and various natural languages. In this work, we consider how generative models of visual data can inherit such benefits. Whereas LLMs have text tokens, Sora has visual patches. Patches have previously been shown to be an effective representation for models of visual data.15,16,17,18 We find that patches are a highly-scalable and effective representation for training generative models on diverse types of videos and images.

At a high level, we turn videos into patches by first compressing videos into a lower-dimensional latent space,19 and subsequently decomposing the representation into spacetime patches.

Video compression network

We train a network that reduces the dimensionality of visual data.20 This network takes raw video as input and outputs a latent representation that is compressed both temporally and spatially. Sora is trained on and subsequently generates videos within this compressed latent space. We also train a corresponding decoder model that maps generated latents back to pixel space. 

Spacetime Latent Patches

Given a compressed input video, we extract a sequence of spacetime patches which act as transformer tokens. This scheme works for images too since images are just videos with a single frame. Our patch-based representation enables Sora to train on videos and images of variable resolutions, durations and aspect ratios. At inference time, we can control the size of generated videos by arranging randomly-initialized patches in an appropriately-sized grid.

Scaling transformers for video generation

Sora is a diffusion model21,22,23,24,25; given input noisy patches (and conditioning information like text prompts), it’s trained to predict the original “clean” patches. Importantly, Sora is a diffusion transformer.26 Transformers have demonstrated remarkable scaling properties across a variety of domains, including language modeling,13,14 computer vision,15,16,17,18 and image generation.27,28,29

In this work, we find that diffusion transformers scale effectively as video models as well. Below, we show a comparison of video samples with fixed seeds and inputs as training progresses. Sample quality improves markedly as training compute increases. 

Variable durations, resolutions, aspect ratios

Past approaches to image and video generation typically resize, crop or trim videos to a standard size – e.g., 4 second videos at 256x256 resolution. We find that instead training on data at its native size provides several benefits.

Sampling flexibility

Sora can sample widescreen 1920x1080p videos, vertical 1080x1920 videos and everything inbetween. This lets Sora create content for different devices directly at their native aspect ratios. It also lets us quickly prototype content at lower sizes before generating at full resolution—all with the same model.

Improved framing and composition

We empirically find that training on videos at their native aspect ratios improves composition and framing. We compare Sora against a version of our model that crops all training videos to be square, which is common practice when training generative models. The model trained on square crops (left) sometimes generates videos where the subject is only partially in view. In comparison, videos from Sora (right)s have improved framing.

Language understanding

Training text-to-video generation systems requires a large amount of videos with corresponding text captions. We apply the re-captioning technique introduced in DALL·E 330 to videos. We first train a highly descriptive captioner model and then use it to produce text captions for all videos in our training set. We find that training on highly descriptive video captions improves text fidelity as well as the overall quality of videos.

Similar to DALL·E 3, we also leverage GPT to turn short user prompts into longer detailed captions that are sent to the video model. This enables Sora to generate high quality videos that accurately follow user prompts.

Animating DALL·E images

Sora is capable of generating videos provided an image and prompt as input. Below we show example videos generated based on DALL·E 231 and DALL·E 330 images.

Prompting with images and videos

All of the results above and in our landing page show text-to-video samples. But Sora can also be prompted with other inputs, such as pre-existing images or video. This capability enables Sora to perform a wide range of image and video editing tasks—creating perfectly looping video, animating static images, extending videos forwards or backwards in time, etc.

Extending generated videos

Sora is also capable of extending videos, either forward or backward in time. Below are four videos that were all extended backward in time starting from a segment of a generated video. As a result, each of the four videos starts different from the others, yet all four videos lead to the same ending.

We can use this method to extend a video both forward and backward to produce a seamless infinite loop.

Video-to-video editing

Diffusion models have enabled a plethora of methods for editing images and videos from text prompts. Below we apply one of these methods, SDEdit,32 to Sora. This technique enables Sora to transform the styles and environments of input videos zero-shot.

Connecting videos

We can also use Sora to gradually interpolate between two input videos, creating seamless transitions between videos with entirely different subjects and scene compositions. In the examples below, the videos in the center interpolate between the corresponding videos on the left and right.

 Image generation capabilities

Sora is also capable of generating images. We do this by arranging patches of Gaussian noise in a spatial grid with a temporal extent of one frame. The model can generate images of variable sizes—up to 2048x2048 resolution.

Emerging simulation capabilities

We find that video models exhibit a number of interesting emergent capabilities when trained at scale. These capabilities enable Sora to simulate some aspects of people, animals and environments from the physical world. These properties emerge without any explicit inductive biases for 3D, objects, etc.—they are purely phenomena of scale.

1.     3D consistency. Sora can generate videos with dynamic camera motion. As the camera shifts and rotates, people and scene elements move consistently through three-dimensional space.

2.     Long-range coherence and object permanence. A significant challenge for video generation systems has been maintaining temporal consistency when sampling long videos. We find that Sora is often, though not always, able to effectively model both short- and long-range dependencies. For example, our model can persist people, animals and objects even when they are occluded or leave the frame. Likewise, it can generate multiple shots of the same character in a single sample, maintaining their appearance throughout the video.

3.     Interacting with the world. Sora can sometimes simulate actions that affect the state of the world in simple ways. For example, a painter can leave new strokes along a canvas that persist over time, or a man can eat a burger and leave bite marks.

4.     Simulating digital worlds. Sora is also able to simulate artificial processes–one example is video games. Sora can simultaneously control the player in Minecraft with a basic policy while also rendering the world and its dynamics in high fidelity. These capabilities can be elicited zero-shot by prompting Sora with captions mentioning “Minecraft.”

These capabilities suggest that continued scaling of video models is a promising path towards the development of highly-capable simulators of the physical and digital world, and the objects, animals and people that live within them.

Discussion

Sora currently exhibits numerous limitations as a simulator. For example, it does not accurately model the physics of many basic interactions, like glass shattering. Other interactions, like eating food, do not always yield correct changes in object state. We enumerate other common failure modes of the model—such as incoherencies that develop in long duration samples or spontaneous appearances of objects—in our landing page.

We believe the capabilities Sora has today demonstrate that continued scaling of video models is a promising path towards the development of capable simulators of the physical and digital world, and the objects, animals and people that live within them.