Introducing Starchild-1 from Odyssey, the first ever real-time multimodal world model. This a model that can generate interactive simulations of the world that you can—for the first time ever—hear. Starchild-1 represents a big step towards a general-purpose world simulator.

@odysseyml can it simulate my mom telling me she’s proud of me ???

@odysseyml If it can't then I want to let you know I'm proud of you.

Starchild-1 is an early step beyond world models that learn only from visual observation, toward systems that learn from richer multimodal interaction with the world.

My dream is that models like Starchild-1 unlock entirely new forms of education, gaming, companionship, robotics, and brand new computing devices. We're so early on this journey, but I'm so excited. The team did truly great work here.

We're also releasing an accompanying technical report of Starchild-1, to share our learnings and encourage further research in this direction!

@odysseyml I need a first ever real-time multimodal world model of my printer

@odysseyml literally it can replace all education schools and online courses faceless era of teachers is here and the level of works which you shared, haven't seen such as once even in Seedance good job

@odysseyml I like to think that one day soon we'll all have Einstein in our pocket, ready to teach us anything in an natural, effective way.

@0x_kaize @odysseyml This could be huge for robotics, gaming, and education. I’m really looking forward to it

@0x_kaize @odysseyml Exactly, and us too!

@odysseyml I heard you liked living in a simulation so I made a simulation generator inside of your simulation. WILD stuff.

@odysseyml What layer of the simulation do you think we're in?

@odysseyml the turtle one

@odysseyml damn this is exciting always use to thought how this will work in real time

@odysseyml congrats on the launch @olivercameron, looks really promising. Gonna give the paper a read

@odysseyml LMK what you think.

@odysseyml Congrats to the entire Odyssey team, this is historic.

@odysseyml really excited to see where this goes, the paper summarises it perfectly. congrats team!

@odysseyml Why thank you!

@odysseyml They’ve done it again, good stuff team 👏

@odysseyml this is pretty cool

@odysseyml It's a strange new form of intelligence. Hard to put it into words.

@odysseyml a serious technical step, as audio forces the model to learn hidden physical and social structure that silent video can often fake. So many possibilities I can think of right away.

@odysseyml This is simply amazing. Congrats to you and the team for such an accomplishment!

@odysseyml This is starting to feel like the future of simulation itself.

@odysseyml Let's be honest we're all in a simulation anyway.

@odysseyml 🔥

@odysseyml Congrats, @olivercameron ! Would love to see Startchild and Odyssey at the upcoming SIGGRAPH hack!

@odysseyml So excited to see this work come to fruition. Incorporating audio into causal world models is an incredibly complex problem and the team did amazing work here. Once you have heard your world models speak there is no going back!

@odysseyml Couldn't agree more.

@odysseyml its crazy how good this is for being the first to do it can’t wait to try it guys!

@odysseyml Starchild-1 is honestly crazy to watch!

@odysseyml whoa 😳 world building just got an upgrade

@odysseyml Exactly.

@odysseyml The future is legit gonna be so freaking insane

@odysseyml Congrats man @olivercameron Let me go check it out

@odysseyml i think in a few years all video generation being done by ai will be done in a live feed way like this

@odysseyml early days... but this feels like the bridge to truly immersive AI.

@odysseyml what a great video, more AI demos need to be like this! showcase the people and their reactions to the tech

@odysseyml How long until they realise that they live in a simulation?

@odysseyml How long until we realize?

@odysseyml Don't do this to me. I haven't figured out how to not think about this 🙏

@odysseyml feels like we just skipped a few years ahead with this one

@odysseyml So it’s like Higgsfield with voice mode plus a big eyeball?

@odysseyml No no. Way different. Starchild-1 generates an interactive audio-visual simulation in real-time. Traditional video models can't do that.

@odysseyml founded by ex-self-driving engineers from Cruise and Wayve, co-founder of Pixar on the board, and their CTO already called this 'the GPT-2 moment for world models' if the scaling laws hold like they did for LLMs, this gets scary fast

@odysseyml I’d like to make a synthetic agent using this. But I need to connect to my data and agentic system. Can I do that?

@odysseyml Say more?

@odysseyml Knowledge that could be in a closed database, file, etc. let’s say you have the synthetic human, which is in the world model. It should have a memory base attached to that single individual or item, guess it could be an animal or object. Allowing recall of specific information.

@odysseyml So is this a simulation inside a simulation?!

Devs has become real… or devs was always real and we are not? Or the simulation in devs is real but the non simulation in devs is not real and the viewer is the simulator? Or the viewer is a simulation and Devs is real and the simulation in Devs is not real? Error… Anyways, very cool!

@odysseyml Devs was a total inspiration for Starchild-1.

@odysseyml Very cool and good luck, excited to try it!

@odysseyml can it recreate what it’d feel like for my girlfriend and me to explore Mars?

@odysseyml Yes absolutely. We built this for you.

@Scobleizer @odysseyml Awesome 👏