A research and development studio building world models for robotics.https://capeki.coenhttps://capeki.co/blog/waypoint-1-5https://capeki.co/blog/waypoint-1-5Higher-fidelity generative worlds are now accessible across consumer hardware with streaming and local execution.Thu, 09 Apr 2026 16:59:00 GMThttps://capeki.co/blog/raising-a-biome-the-trials-and-tribulations-of-waypoint-at-homehttps://capeki.co/blog/raising-a-biome-the-trials-and-tribulations-of-waypoint-at-homeAt Overworld, we want you to play our open-weight world models on your very own computer, and Biome is our solution: a cross-platform, fully-integrated, open-source client. But how'd we build it? And what challenges did we run into?Fri, 13 Mar 2026 18:04:56 GMThttps://capeki.co/blog/engineering-safety-for-interactive-world-modelshttps://capeki.co/blog/engineering-safety-for-interactive-world-modelsSafety at Overworld starts well before a prompt reaches the model. From dataset filtering to runtime safeguards and open releases, we’ve been building a layered safety system across the entire pipeline. This post is not intended to weigh in on broader moral or philosophical debates surrounding AI.Fri, 06 Mar 2026 23:34:39 GMThttps://capeki.co/blog/transforming-prompt-to-worldshttps://capeki.co/blog/transforming-prompt-to-worldsOur prompting pipeline rewrites user input into structured signals the world model actually understands: seed images, synthetic video, and aligned controls. It started as a safety filter. It became the other half of the product.Fri, 20 Feb 2026 18:20:27 GMThttps://capeki.co/blog/ditohttps://capeki.co/blog/ditoGANs are tough. Diffusion is simple. Diffusion VAEs are the next step in controllable HD reconstructions.Fri, 13 Feb 2026 17:00:00 GMThttps://capeki.co/blog/the-immersion-gaphttps://capeki.co/blog/the-immersion-gapIn a race to maximize visual fidelity, the fun factor of world models has suffered. We've arrived in an era where a full second of latency is considered playable, 4fps is considered real-time, and a rack of $50,000 GPUs is considered accessible. We want to fix this—and Project Genie reminds us why it matters.Fri, 30 Jan 2026 18:00:00 GMThttps://capeki.co/blog/the-path-to-real-time-worlds-and-why-it-mattershttps://capeki.co/blog/the-path-to-real-time-worlds-and-why-it-mattersToday we're releasing Waypoint-1, the first real time diffusion world model optimized for consumer GPUs.Tue, 20 Jan 2026 17:47:39 GMThttps://capeki.co/blog/scaling-up-data-collectionhttps://capeki.co/blog/scaling-up-data-collectionWe needed a bunch of annotated game data, we were in a hurry, and we made it happen as fast as we possibly could. It was tricky, we broke a lot of things, and it worked. Here's what we did, what worked and what we broke.Fri, 02 Jan 2026 17:12:46 GMThttps://capeki.co/blog/world-model-optimization-quantizationhttps://capeki.co/blog/world-model-optimization-quantizationAt its heart, the post tackles a critical bottleneck in large-scale transformer-based generative models: the KV cache. During inference, this cache stores the context from previous steps, but it can grow very large, consuming memory and hitting bandwidth limits as data is repeatedly read by the GPU. In this blog post, we detail how we address this via quantization and other optimization techniques.Sat, 13 Sep 2025 19:49:00 GMThttps://capeki.co/blog/why-human-evaluation-is-the-missing-piece-in-world-model-developmenthttps://capeki.co/blog/why-human-evaluation-is-the-missing-piece-in-world-model-developmentWe're releasing OWL Eval, the first open-source evaluation platform built specifically for studying how humans perceive AI-generated videos. After running studies with hundreds of participants, we've learned that human evaluation reveals critical model failures that automated metrics completely miss. Our platform makes it dead simple to run these studies at scale.Fri, 01 Aug 2025 17:17:11 GMThttps://capeki.co/blog/diffdec-ode-regressionhttps://capeki.co/blog/diffdec-ode-regressionWe show applying ODE regression to drastically reduce the depth of our diffusion decoder, leading to a 40x speedup!Sun, 27 Jul 2025 02:06:00 GMThttps://capeki.co/blog/product-of-experts-for-visual-generationhttps://capeki.co/blog/product-of-experts-for-visual-generationIn this blog post, we illustrate a paper that leverages multiple specialist models and incorporating their individual expertise by having them influence the diffusion sampling at inference time. We also provide code examples, visualizations, and intuitions!Sat, 12 Jul 2025 00:02:10 GMThttps://capeki.co/blog/fast-audio-video-world-models-part-2https://capeki.co/blog/fast-audio-video-world-models-part-2We trained an autoencoder with depth maps in the latent. It resulted in far better depth consistency in downstream generations. Next we’re training with optical flow as well, and solving the KV cache problemWed, 02 Jul 2025 00:05:29 GMThttps://capeki.co/blog/gen-vs-rechttps://capeki.co/blog/gen-vs-recThe generation vs reconstruction trade-off gets weird when you push compression. Learn more about how we're managing it in this blog post!Sat, 21 Jun 2025 00:46:08 GMThttps://capeki.co/blog/fast-av-wms-part-1https://capeki.co/blog/fast-av-wms-part-1Fri, 13 Jun 2025 22:47:49 GMThttps://capeki.co/blog/training-diffusable-autoencodershttps://capeki.co/blog/training-diffusable-autoencodersJoin us as we try to figure out how to make a good custom autoencoder for our World Model.Sat, 07 Jun 2025 00:10:00 GMThttps://capeki.co/blog/training-idmshttps://capeki.co/blog/training-idmsThis week we set our sights on taming unlabeled internet data for World Model training.Fri, 30 May 2025 19:38:00 GMThttps://capeki.co/blog/towards-a-large-open-video-game-datasethttps://capeki.co/blog/towards-a-large-open-video-game-datasetToday we are marking the start of our journey towards a general purpose open source video game world model.Fri, 23 May 2025 23:59:09 GMT