Show HN: I built a playable 211M-param world model of the Backrooms A developer released a 211M-parameter transformer world model, called noclip, that generates each next frame of a playable Backrooms game from the previous 2.4 seconds of frames and player inputs, with weights published on Hugging Face. The model was trained from scratch in PyTorch on 440 hours of scripted play recorded across a simulator of 18 Backrooms levels, using flow matching in a pretrained autoencoder's latent space, and was fed its own imperfect frames as context to limit drift. Running it requires an NVIDIA GPU with roughly 4 GB of memory; the developer cited Oasis Minecraft as inspiration. A 211M-param transformer acts as a game engine. It draws every next frame from the last 2.4 seconds and my inputs. Built a simulator of 18 Backrooms levels in PyTorch to generate the data. Recorded 440 hours of scripted play in it. Trained the model from scratch on that, using flow matching in a pretrained autoencoder's latent space. The model was trained on its own imperfect frames as context so it doesn't drift right away. It recalls a few older frames by position, so a quick look back usually works. Inspired by Oasis Minecraft. To try it you need an NVIDIA GPU with ~4 GB. Weights: https://huggingface.co/yelarys/noclip https://huggingface.co/yelarys/noclip Video with sound: https://x.com/yelaryss https://x.com/yelaryss Comments URL: https://news.ycombinator.com/item?id=50594070 https://news.ycombinator.com/item?id=50594070 Points: 1 Comments: 0