Company profile · 3 min read
Lucid: interactive video world models
Lucid builds action-conditioned video models for interactive worlds, with research applications in gaming and robotics simulation.
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
Published · Updated
What it does
Lucid is building interactive video world models: neural environments that generate frames in response to actions instead of relying on a hand-coded game or physics engine. The YC profile describes a Minecraft simulator running at 20+ frames per second on an RTX 4090, with the longer-term goal of training robots and generating open-ended game worlds. The current homepage is intentionally sparse, so the YC launch remains the substantive public product description checked here.
The fit is a robotics, embodied-AI or game-research team that needs a learned simulator rather than another static 3D environment. Lucid’s promise depends on the model preserving useful dynamics under control and transferring beyond the training footage. A fast demo is not enough if the world model cannot support a repeatable task or a safe sim-to-real evaluation.
Why I’d look closer
The technical wedge is unusually clear. Lucid says it trained a neural Minecraft simulation from 200 hours of gameplay, reaching 20+ FPS and roughly five times the speed of other Minecraft world models, while using aggressive tokenizer compression and 100x fewer resources. Those are company-reported technical claims. The product direction also matters: the launch says early partners can fine-tune domain-specific LoRAs for robotics or other high-fidelity simulation needs.
The founder context is hands-on. The YC profile describes Rami Seid as a former machine-learning engineer at a robotics lab, later CTO of a govtech contractor and co-founder of a telecommunications company. The launch post also names Alberto as a co-founder, while the current directory’s active founder section lists Rami; that roster detail should be kept separate from the product thesis.
What I’d ask
What actions and environments are supported beyond Minecraft, how is visual fidelity measured, and where does the simulator fail under distribution shift? I’d ask for a fixed task suite, latency/quality tradeoff, data licensing, controllability tests and a sim-to-real plan for robotics. No public pricing or current partner list was exposed in the sources checked.
My editorial take
Lucid is worth watching for teams that see simulation as the bottleneck to embodied AI. The speed claim is a useful signal, but the decision should turn on controllability and transfer, not generated-video quality in isolation.
Quick facts
| Field | Sourced detail |
|---|---|
| Buyer fit | Robotics, embodied-AI and interactive-world research teams |
| Product | Action-conditioned video world models |
| Technical signal | Company reports 20+ FPS Minecraft simulation and 5x speed versus other models |
| Public pricing | Not exposed in the sources checked |
Sources checked
Checked 2026-09-19.
| Source | Used for |
|---|---|
| YC company profile | Product, founders and technical claims |
| Lucid homepage | Current public status surface |
Cohort context
Lucid is listed in Winter 2025. In our 2026-09-18 directory snapshot, 8 of 165 listed companies in that cohort have YC’s primary industry label Consumer (4.8%). This is a current-directory comparison, not an original intake count or a performance ranking. Nine-cohort dataset.
Public website snapshot
Observed 2026-09-19T16:19:38.440Z in raw homepage HTML. This records visible metadata and advertised links, not agent execution or product quality.
| Signal | Homepage observation |
|---|---|
| Product description metadata | Observed |
| Canonical link | Not observed in this response |
| H1 or H2 heading | Observed |
| Typed structured data | Not observed in this response |
| Docs/developer link | Not observed in this response |
| Pricing link | Not observed in this response |
| llms.txt link | Not observed in this response |
| Markdown alternate | Not observed in this response |
Public observations · Collection method. Missing links here do not establish that a capability or file is absent elsewhere.
About the author
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.
First1000 ↗ · X ↗