Odyssey

实时生成可交互3D视频世界的AI世界模型

video generation

https://odyssey.ml/

Visit Website

In-depth Report

Odyssey is a revolutionary world model product developed by London-based AI startup Odyssey AI Lab, which can generate interactive 3D video worlds in real time. Unlike traditional video generation models, Odyssey allows users to "walk into" video content for a real-time interactive experience just like playing a game. The company, founded by self-driving pioneers Oliver Cameron and Jeff Hawke, has raised more than $27 million in funding from investors including EQT Ventures, GV (Google Ventures) and Pixar co-founder Ed Catmull. At the technical level, Odyssey uses a frame-by-frame prediction mechanism to generate a frame every 40 milliseconds, achieving real-time performance of starting streaming output within 50 milliseconds. Although the current product is still in its early stages and the visual quality and stability need to be improved, the direction of the new generation of interactive video technology it represents deserves attention.
Odyssey is a London-based artificial intelligence startup founded around 2023 by two senior experts in the field of autonomous driving. Co-founder Oliver Cameron previously held an important position in self-driving AI research at Wayve, while Jeff Hawke was the CEO of Voyage self-driving company. The two successfully grafted the "world modeling" concept in the field of autonomous driving into AI video technology, creating Odyssey's unique real-time interactive video generation capabilities. The company’s board of directors has a luxurious lineup, including Ed Catmull, co-founder of Pixar and former president of Disney Animation Studios, who brings a deep background in the entertainment industry and experience in creative content production to the company. This personnel arrangement clearly shows Odyssey’s ambition to not only be a technology tool, but also to become the next generation entertainment content platform. At the capital level, Odyssey has completed more than $27 million in financing, with major investors including European first-tier fund EQT Ventures, Google Ventures GV and Samsung Next. The endorsement of these top investment institutions not only provides the company with ample R&D funds, but also brings it abundant industrial resources and strategic support.
The core function of Odyssey is to transform static videos into an interactive, real-time 3D world. Users can control the perspective through keyboards, controllers and other devices, and freely explore the generated content. The most prominent technical feature of the product is its real-time nature: a frame is generated every 40 milliseconds, and the user can see the response in only 50 milliseconds after input, achieving a truly real-time interactive experience. Unlike traditional video generation models that require minutes of waiting, Odyssey allows users to start exploring the video world instantly. In terms of output duration, Odyssey is able to generate consistent video streams of 5 minutes or more, breaking through the limitation of traditional AI video models that can only generate 10-second clips. The system supports two input methods: image prompts and text prompts. Users can generate interactive worlds based on existing images or text descriptions. From the perspective of technical architecture, Odyssey adopts the world model route, which is in line with the "world model" concept advocated by authorities in the AI field such as Yann LeCun and Li Feifei. Different from the frame-by-frame generation of traditional diffusion models, Odyssey simulates the idea of predicting the next word from a large language model, continuously predicting the content of the next frame, and forming a coherent world evolution. In order to solve the common "picture drift" problem in the field of AI videos - that is, the gradual distortion and deformation of the picture over time - Odyssey adopts a "narrow-field pre-training" strategy to first cultivate basic understanding on a large number of general videos, and then fine-tune it for specific environments to improve stability. In terms of product positioning, Odyssey has made it clear that it will work with creative professionals rather than replace them. The software allows the generated scenes to be exported to professional tools such as Unreal Engine, Blender, Adobe After Effects, etc. for manual editing. This positioning can help alleviate the anxiety of the creative industry about AI replacing humans. The current product is still in its early stages and has obvious limitations. The visual quality of the generated environment is not clear enough, with a certain degree of blur and distortion; the spatial consistency is insufficient, and the surrounding environment may change significantly after walking for a period of time or turning around in the scene; the temporal stability needs to be improved. These "rough edges" are a reality acknowledged by the company itself, which says it is making improvements in aspects such as richer world representation, improved temporal stability, and expanded action space.
As of now, Odyssey is still in the early free trial stage, and users can try it out through the official website experience portal. However, due to limited GPU supply, trial quota may be limited. For the commercialization path, the company has opened API access applications, and developers can obtain API interfaces through the Odyssey developer platform. The official website shows that the operating cost is approximately US$1-2 per user hour (based on Nvidia H100 GPU cluster operation). This cost structure determines that future commercial pricing needs to be adjusted based on this pricing. From a business model perspective, Odyssey’s vision is to “convert all video content into interactive videos.” This positioning means that its commercialization path may include: API call payment for enterprises, professional version subscriptions for creators, and an entertainment content platform for consumers. Given the team’s strong background in autonomous driving, the company may also license the technology to application scenarios in fields such as robotics and intelligent navigation.
Judging from technical demonstrations and early user experience feedback, Odyssey has attracted a lot of attention. Users are generally excited about its innovative interactive form of "walking into video", and many technology media have carried out special reports. The ability to interact in real time is considered a revolutionary breakthrough, and the 50 millisecond response delay makes an "instant start" experience possible. However, early experiences have revealed some criticism. The core problem focuses on visual quality - the generated environment is reported to be "blurry and distorted", which is far from the clarity of current top AI video tools (such as Runway, Pika). In addition, scene stability is another frequently mentioned issue. Some users report that after moving in the scene, the surrounding environment will "look suddenly different". Judging from social media discussions, technology enthusiasts and AI practitioners generally have a positive attitude towards Odyssey, believing that it represents a new technology paradigm. But feedback from the creative community is more complicated: On the one hand, some people expect this tool to improve creative efficiency, but on the other hand, they are worried about whether it will further impact creative jobs that are already affected by AI.
Odyssey has received a lot of attention from mainstream tech media. TechCrunch describes it as an innovative product that "allows users to interact with streaming videos" and believes that it creates a new concept of "interactive video". Industry observers generally rank Odyssey alongside DeepMind, World Labs (founded by AI research pioneer Li Feifei), Microsoft and other companies that are developing world models, believing that this technical direction is becoming the next hot spot in the field of AI video. Looking at the competitive landscape, Odyssey isn't the only company chasing the world model. DeepMind has strong R&D resources and talent reserves, World Labs is backed by top academic background in the field of AI, Microsoft has released a demonstration of AI-generated "Quake 2", and Decart has developed an AI simulation of "Minecraft" that can be played in real time. In this highly competitive track, Odyssey’s differentiated advantage lies in the real-time interaction capabilities brought by its founding team’s self-driving background, and the entertainment industry perspective brought by Pixar’s background. Expert analysis believes that the maturity of world model technology will profoundly change the way content is produced. Industry predictions indicate that in the future this technology can be used to create interactive media (games, movies), run real simulations (such as robot training environments), as well as in entertainment, advertising, education, training, tourism and other fields. Whether Odyssey can stand out from the competition depends on its technology iteration speed and commercialization capabilities.
The biggest controversy Odyssey faces comes from concerns in the creative industry about AI replacing humans. Game studios such as Activision Blizzard, which is already using AI technology to cut costs and combat attrition, have seen massive layoffs, a Wired investigation found. A 2024 study by the American Animation Society estimates that more than 100,000 U.S. film, television and animation jobs will be impacted by AI in the coming months. Although Odyssey states that it is positioned to "collaborate with creative professionals" rather than replace it, the neutral nature of the technology itself leaves uncertainty about how it will be applied. If this technology is used on a large scale to reduce content production costs, it may intensify employment pressure in the creative industry. From a technical risk perspective, the lack of maturity of current products is the main problem. Core indicators such as visual quality, spatial consistency, and temporal stability are still far from commercial application standards. Companies need to continually invest R&D resources to make improvements, a process that can take years. Furthermore, as an emerging technology, the regulatory framework for world models is unclear. If AI-generated content involves risks such as copyright issues and the spread of false information, it may face uncertainty caused by policy adjustments.
Odyssey is currently most suitable for the following user groups: AI technology researchers and developers can access through API to explore the technical boundaries of the world model; creative content creators can use its generation capabilities as a source of inspiration or drafting tool, and export to professional software for refined editing; technology enthusiasts can experience the cutting-edge concepts of interactive video. For ordinary consumers, the free experience at this stage is worth a try, but expectations need to be managed - the product is still in its early stages, and the visual effects and stability are far behind mature products. Scenarios that are not suitable for using Odyssey at the current stage include: professional production teams that require high-quality commercial video content (it is recommended to wait until the technology matures before evaluating); application scenarios that require extremely high interaction stability; and applications that have strict requirements for picture clarity. In terms of alternatives, if the need is traditional AI video generation, you can consider mature products such as Runway, Pika Labs, and Luma Dream Machine; if you need 3D scene generation, related products from World Labs and Microsoft are worthy of attention; if the goal is real-time AI generation in the game, Decart's Minecraft AI simulation is in a similar direction.
Odyssey represents an important technological breakthrough in the field of AI video generation - moving from passive viewing to active interaction. Although the current product is still in the early stages of "rough edges", its real-time streaming output and open interaction capabilities give us a glimpse of the next generation of content formats. The company's founding team has both technical depth and industrial vision, sufficient financing and the blessing of top consultants, laying the foundation for long-term development. However, employment anxiety in the creative industry, challenges with technological maturity, and increasingly fierce market competition are all risks that Odyssey needs to face up to. For practitioners and enthusiasts who are concerned about the direction of AI content technology, Odyssey is a project worthy of continued attention. However, when its promised vision of "video content interactivity" can be truly popularized, it remains to be seen how technology will evolve and the market will be tested.

User Reviews

Frances.Wilson_2023

—

刚体验了一波 Odyssey，不得不说这个实时交互是真的香！50毫秒响应，玩起来基本感觉不到延迟，比之前用过的那些AI视频工具流畅太多了。

Judy.Castillo

—

和 Runway、Pika 比起来，Odyssey 的交互性确实是独一档，但画质还是有点拉胯，生成的场景有些模糊，希望后续能优化。

暖阳574

—

看演示视频觉得挺牛的，但实际用起来发现画面漂移问题还是存在，走一段时间后周围环境就变样了，稳定性有待提高。

trueJohnniNewman_2024

—

创始人是从 Wayve 出来的自动驾驶大牛，团队还有皮克斯的 Ed Catmull 坐镇，这背景是真的强，融资2700万美元不奇怪。

Cha_inDex

—

免费的体验名额太少了，GPU 供应有限，每次想用都要排队等，哭了。

Nancy_Scott369211

—

技术方向挺看好的，世界模型这条路线比传统扩散模型更有想象力，期待后续版本的表现。

jaSUL

—

用键盘 WASD 控制就能在生成的视频里漫游，有一种玩第一人称游戏的感觉，沉浸感拉满了。

realHarveyBlack_pro

—

40毫秒生成一帧是什么概念？人类眨眼都要100毫秒，AI这波真的是瞬间生成了，响应速度快到离谱。

Andrew.Rogers007

—

目前还在早期阶段，demo 确实比较粗糙，边缘有些模糊，但对于一个新领域来说已经很强了。

AnthonyNielsen

—

可以导出到 Unreal Engine、Blender 这些专业工具进行二次编辑，这点对创作者很友好。

NObai

—

刚在 Product Hunt 上看到 Odyssey-2 Max 上榜了，拿到了 123 个赞，确实很受关注。

3yc_97u

—

和美国那些世界模型项目比如 DeepMind、World Labs 相比，Odyssey 的实时交互能力是最大亮点。

Elizabeth.KellyQ818

—

能生成5分钟以上的长视频流，这点比只能生成10秒的竞品强太多了，场景持续性更好。

JohnCox520

—

支持图像和文本两种提示方式，用起来比较灵活，我比较喜欢用图像生成。

Jacqueline.Vasquez16830

—

运营成本每用户小时1-2美元，未来如果收费的话这个价格还算合理，就看体验值不值这个价了。

Henry_ThompsonK

—

看到知乎上有人讨论说这个可能会冲击创意产业岗位，确实有点担忧，做视频的要失业了？

AKelly_2022

—

用了窄域预训练策略来解决画面漂移，思路挺聪明的，实际效果也还行。

IMyers_20231

—

说实话现在这画质还达不到商业应用的标准也就是玩票性质但技术潜力很大值得持续关注。

Brandon.HowardSr4

—

看了 TechCrunch 的报道，说这技术可以应用于游戏、电影、教育、旅游等多个领域，前景广阔。

SamanthaMurphy

—

两位创始人一个来自 Wayve，一个来自 Voyage，都是自动驾驶领域的，做的却是视频生成，跨界的思路很妙。

Odyssey

In-depth Report

User Reviews

Similar Products