DreamForge-World 0.1 Preview reaches 15 FPS interactive simulation on single RTX 4090
Researchers release a preview world model that runs real-time interactive simulations at 480p on consumer GPUs, adapting open video backbones for keyboard and mouse control with minute-scale rollouts.
DreamForge-World 0.1 Preview, a foundational world model from researchers Daniyel Ayupov and Artur Markov-Tsoy, runs real-time interactive simulations on consumer hardware. The system adapts the LongLive 1 autoregressive video stack—itself derived from Wan2.1-T2V-1.3B—and adds a residual action pathway inspired by the Matrix-Game family. It supports live keyboard and mouse control, multimodal initialization, mid-stream reprompting, dual-view operation, and minute-scale interactive rollouts at native 480p resolution, reaching 14 to 15 FPS on a single RTX 4090 with a low memory footprint.
The project targets a different axis than frontier-scale world simulators. Instead of chasing maximum fidelity or memory completeness, DreamForge-World 0.1 Preview prioritizes low-compute adaptation and broad interactive capability coverage. The authors built the preview system by leveraging open video backbones and applying targeted adaptation runs, achieving high cost-efficiency.
Consumer GPU performance
The system runs on a single RTX 4090 at 14 to 15 FPS, a frame rate that makes real-time interaction practical for local experimentation. The 480p native resolution and minute-scale rollouts mean users can explore interactive scenarios without cloud infrastructure or multi-GPU clusters. The dual-view operation and mid-stream reprompting features let practitioners adjust scenes on the fly, a capability that frontier models typically gate behind API endpoints.
The authors acknowledge that DF-World 0.1 Preview is not yet a memory-complete or frontier-quality world simulator. The focus is on demonstrating a practical low-compute route toward real-time controllable world-model previews on consumer GPUs. For practitioners running local setups, the combination of open video backbones, interactive control, and single-GPU runtime makes this a testable preview of where consumer-hardware world models might land in the near term.



