Our spatial memory system helps video world models reconstruct familiar places.
94x less storage. 44ms retrieval from a 1.65-hour archive.
Learn more ↓
research.decart.ai/spatial-m…
What if an AI model could take back a harmful answer before it finishes it?
Diffusion language models can.
New Decart research accepted to #EMNLP2026 found harmful responses in just 9–21% of attacks, vs. 48–62% for comparable autoregressive models.
We turned that behavior into SRI Guard - detecting harmful generations with just 0.01–0.04% added latency.
Safety shouldn’t come at the cost of speed.
Read the paper: research.decart.ai/step-wise…
@INSIGHTLabBGU @TechnionLive @bengurionu @CS_Uni_of_Haifa
How much can we optimize?
Quite a lot, it turns out.
Decart Research is now live - a closer look at the work making our models faster, more efficient, and more capable.
Explore our publications, programs, and tools for the global research community - with more to come.
🌐 research.decart.ai
.@DecartAI founding member @AbermanKfir says world models could let robots train on real-world physics without needing to experience every scenario in the physical world:
"If LLMs can imitate the semantic world, world models can imitate the physical world."
"Imagine that you have a humanoid in a different environment. You'd have to collect this impossible type of data."
"But if you have world models that can simulate it with real physics, you'll be able to train these robots."
"That's the premise where we're at now. It hasn't been resolved yet."
Okay third time is a charm & real-time & in a live app! App & code in 🧵.
Real-time video gen is getting better than I thought. Thanks @nickjohn @meh_agarwal for recommending @DecartAI Lucy!
Check out Anywear!
Great example of real products that are only possible with Live AI. This time, built for e-commerce!
1/ We’re entering the era of agentic commerce, and we believe that world models are the missing engine.
Instead of browsing static product pages, agents will let you interact with products inside your own world before making decisions.
Today, @DecartAI we're releasing Anywear👇
Decart retweeted
Okay third time is a charm & real-time & in a live app! App & code in 🧵.
Real-time video gen is getting better than I thought. Thanks @nickjohn @meh_agarwal for recommending @DecartAI Lucy!
Decart retweeted
Damn this is so cool.
Finger frame your face and transform into different styles in real-time ✨
App made by @sophiamyang, powered by @DecartAI
Decart retweeted
I’ve been building LucyLive — a real-time AI render plugin for @MaxonVFX Cinema 4D.
uses @DecartAI Lucy 2.5 through @fal
It follows your prompts and reference images while keeping C4D responsive.
Free download:
github.com/VantageInc/LucyLi…
I’d love to see your results.
Decart retweeted
Earlier this month we hosted the first-ever Vintage Enterprise AI Summit in Paris 🇫🇷✨
A huge thank you to our partners @UnframeAI, @alice_dot_io, @DecartAI, @HappyRobot, @USTglobal, and to @RaiseSummit , @amarmic, and @HenriDelahaye for bringing such an exceptional global AI community together in Paris. It was an honor to partner with you and build this alongside the broader RAISE week 🥂
This is Value+ in action: bringing the right founders, enterprises, investors, and LPs together, and turning those connections into lasting value.
Look out for the next edition! 👀
We’ve got the physics going too!
Replying to @bobdoylemedia
The obvious use case is becoming a character on camera.
But the bigger deal is prompt-controlled transformation: characters, outfits, environments, objects, and visual effects changing while the camera is still live.
Decart retweeted
1/ Today, we’re excited to introduce Lucy 2.5.
Lucy represents a paradigm shift in world models, not just in how generated worlds look, but in how we interact with them.
A thread on what makes it a paradigm shift 👇
exactly the point 🔥
waiting to hear your thoughts on it! im sure you'll be surprised 🔥
Replying to @DecartAI
Realtime editing raises
a very different challenge.
A polished render can hide mistakes.
A live stream cannot.
I’d watch how Lucy handles
hands,
reflections,
and sudden camera movement
without breaking the illusion.
That is where I’d judge the model :)
Decart retweeted
Meet live AI
Meet Lucy 2.5
Meet the most advanced live AI model!
just tried it, result is cool...
🤝 Paid partnership
Decart retweeted
Virtual try-ons and live shopping just got a lot more powerful.
@DecartAI ’s Lucy 2.5 can change outfits, backgrounds, or even entire scenes in realtime while someone is presenting a product.
No more pre-recorded videos or clunky green screens.
This feels like the next step toward truly interactive shopping experiences.
The gap between seeing something online and feeling like you’re trying it on is shrinking fast.
🤝 Paid partnership
Decart retweeted
Real-time AI editing is moving insanely fast. This feels like a glimpse of what’s coming for creators, livestreams, and e-commerce. 👀
🤝 Paid partnership
it's exactly what we made it for! Streamers and Creators just got a great new tool to add to their arsenal
For streamers and live creators, this is huge.
Imagine audience members being able to influence what happens on screen in realtime changing scenes, adding effects, or triggering moments all powered by a live video model that actually understands the request.
@DecartAI ’s Lucy 2.5 makes that kind of interactive streaming feel much closer to reality.


















