What Is Decart AI? The Lab Behind Real Time AI Video

Over the past two years, Decart has raised more than $450 million and drawn attention from major names across the tech industry, most recently with reports that Anthropic is in talks to acquire the company at a valuation of up to $6 billion. Decart was founded in late 2023, so what is this company actually building? This article walks through its main products, Lucy, Oasis, and DOS, and how they work, to give you a clearer picture of the technology behind a name that has caught the attention of both investors and major tech players.

What Is Decart AI?

Decart is an AI lab focused on building world models and video processing models that operate in real time, meaning they respond almost instantly instead of processing content in batches and returning results afterward. The company describes itself as building "infrastructure and models that make AI run at the speed of reality".

Decart's core philosophy centers on speed and computational efficiency. Rather than simply competing to produce better looking AI video, Decart focuses on pushing latency down to just a few dozen milliseconds and cutting computational cost so a model can run continuously, without a fixed duration limit, unlike most other AI video models that can only generate short clips lasting a few seconds. To pull this off, Decart builds technology across the entire technical stack: from optimizing how AI models run on each type of chip, to the video and world models themselves.

Who Is Decart AI For?

Developers and product teams

This is Decart's core user group. Through its API and SDK, developers can integrate real time video editing or world model capabilities into their own applications without building AI infrastructure from scratch.

Retail and e-commerce businesses

Brands can use Decart's technology to build virtual try-on features directly on video, letting customers see how a product looks on them before buying, without waiting for processing.

Advertisers and livestreaming platforms 

With the ability to edit video while it's being broadcast live, ad platforms or livestreaming services can insert products, change backgrounds, or create interactive effects while content is being watched, instead of having to prebuild everything in advance.

People working in robotics and self-driving vehicles

With the Oasis world model, robotics and self-driving vehicle teams can use a simulated environment that responds in real time to train and test systems without deploying them in the real world right away.

Standout Features

Lucy - real time video editing 

Lucy is a model that transforms video while it's happening: changing a character's outfit, adding or removing objects, changing the visual style, or adding special effects, all with latency under 40 milliseconds. That's a major difference from typical AI video editing tools, which need to process footage offline before producing a result.

Oasis - a world model for physical AI

Oasis generates interactive environments that respond in real time with physical accuracy, designed for robotics, self-driving vehicles, manufacturing, and drones. Instead of just producing an illustrative video, Oasis simulates a world where a user's (or an AI system's) actions genuinely affect the surrounding environment in real time.

DOS - AI inference infrastructure optimization

The Decart Optimization Stack is an infrastructure layer that helps AI models run more efficiently across different chip types (Nvidia, AWS Trainium, Google TPU). This is how Decart maintains low latency without massive computational costs. According to the company, efficiency gains can reach up to 100 times over conventional operation.

A continuous video generation architecture with no duration limit

Unlike most AI video models, which can only produce short clips lasting a few seconds, Decart's technology generates video frame by frame, based on the previous frame and the user's prompt, allowing content to run continuously and be edited in real time as it plays, rather than being capped at a fixed length.

Benefits Of Using Decart AI

A smooth experience with almost no lag

For applications that need an instant response, like livestreaming or virtual try-on, latency under 40 milliseconds keeps the experience seamless, and viewers barely notice there's AI processing happening behind the scenes.

No need to build AI infrastructure from scratch

Product teams can use Decart's existing API and SDK instead of having to research, train, and optimize a real time AI model themselves, which is a genuinely complex technical challenge.

Costs scale with actual usage

Because pricing is based on seconds processed or generations produced, small teams can experiment with low upfront costs instead of committing to a fixed subscription from day one.

How To Use It (Overview)

Since Decart delivers its product as models and APIs meant to be integrated into other systems, the actual workflow varies depending on each team's goals. Because of that, the guide below only covers a general overview of the process:

Create an account at platform.decart.ai

This is where you manage your API key, track usage, and handle billing.

Get an API key

After signing up, the platform issues a unique API key, used to authenticate every request sent to Decart's models.

Install the SDK or call the API directly

Decart provides an SDK for JavaScript/Node (npm install @decartai/sdk), or you can call the REST API directly if you're using another language.

Choose the right model for your needs

For example, Lucy for real time video editing, Oasis for an interactive world model, or one of the image editing models.

Send a request with your input content and prompt

For existing video or images, you send along a prompt describing the change you want (for example, "change the character's shirt to red"). For a real time stream (like a video call or webcam feed), the system processes continuously while the stream is running.

Receive and use the output

For batch processing, the result comes back as a processed file. For real time processing, the output is streamed directly so it displays immediately.

If you're not into code but still want to get a feel for how this technology works, Decart also has a few consumer apps, like Mirage or Glint, that let you try it directly through your phone's camera without writing a single line of code.

Tips For Better Results

Pick the right model for the job

Decart offers several model versions at different price points and for different purposes (for example, Lucy 2.5 for high quality edits, Lucy Restyle 2 for simpler style changes at a lower cost). Choosing the right one helps optimize both quality and cost.

Write prompts that describe the specific change, not the whole scene 

Since this is a model that edits existing video rather than generating something from scratch, the most effective prompts tend to be short, clear instructions about what needs to change (for example, "add sunglasses to the person in frame") rather than a full scene description.

Keep a close eye on usage, since pricing is per second

For applications that run continuously in real time, costs can add up quickly if you're not tracking them. It's worth testing with short clips first, estimating real costs, and only then scaling up to full deployment.

Pricing

Decart charges on a pay-as-you-go basis, with no fixed subscription and no minimum spend required:

• Realtime models (Lucy 2.5, Lucy VTON 3.5, Oasis 3 Preview): around $0.02 per second of processing at 720p; Lucy Restyle 2 is around $0.01 per second.

• Batch video processing models: around $0.04 per second at 720p for Lucy 2.5 and Lucy VTON 3.5; around $0.01 per second for Lucy Restyle 2.

• Image editing models (Lucy Image 2): around $0.01 to $0.02 per generation, depending on resolution.

• New accounts receive a batch of free credits to test the API before paying.

• Custom enterprise pricing is available for high volume use, available by contacting [email protected] directly for a quote.

Note: this is pricing built for developers and businesses integrating through the API, not consumer subscription pricing. Prices may change, so it's worth checking the official pricing page before deploying at scale.

Limitations Of Decart AI

Not a tool you can just open and use if you don't code

Most of Decart's features are built for developers integrating through the API, so for people without a technical background, hands-on access is fairly limited, aside from a couple of dedicated consumer apps like Mirage or Glint.

Costs are hard to predict in advance for large scale applications

Since pricing is based on actual seconds used, high traffic applications or ones running continuously in real time can rack up significant costs if usage isn't monitored closely from the start.

Some models are still in preview

Oasis 3, for example, is explicitly labeled as a preview release, meaning the technology is still being refined and may still change or not yet be fully stable.

Documentation and the interface are entirely in English

All of Decart's technical documentation, platform interface, and support channels are currently English only, which could be a barrier for teams not used to working with technical documentation in that language.

Overall

Decart is the technology infrastructure behind a lot of real time AI experiences you may have already seen without knowing who was behind them, from virtual try-on features on video to live scene transformation effects on a stream. What stands out most about Decart is speed: latency of just a few dozen milliseconds, and the ability to generate content continuously with no duration limit, something most other AI video models still can't do. If you're a developer or a product team looking for a real time AI solution, Decart is a name worth digging into further through its official API documentation. If you are looking for an AI solution that delivers fast results, Decart AI is a name worth exploring.