DreamX-World 1.0: A General-Purpose Interactive World Model
DreamX Team, Yancheng Bai, Rui Chen, Xiangxiang Chu, Rujing Dang, Hao Dou, Bingjie Gao, Qiwen Gu, Siyu Hong, Jiachen Lei, Geng Li, Jifan Li, Ruimin Lin, Qingfeng Shi, Bingze Song, Lei Sun, Jing Tang, Ruitian Tian, Jun Wang, Jiahong Wu, Pengfei Zhang, Shen Zhang, Jiashu Zhu
arXiv:2606.16993 · 113▲ · cs.CV
View on arXiv →Premium readers get an interactive explainer for this paper — a figure you can poke at, not just read.
What it is
DreamX-World 1.0 is an interactive text/image-to-video world model built by fine-tuning Wan2.2 to support camera navigation, revisiting earlier scenes, and prompted multi-object events across photorealistic, game, and stylized domains. It combines a camera-conditioning method called E-PRoPE (projective positional encoding applied to spatially downsampled tokens), geometry-based memory retrieval, DMD distillation into a few-step autoregressive generator, and RL alignment, with serving optimizations to hit real-time streaming.
Why it matters
This is an integrated recipe for turning an offline video diffusion model into a streaming interactive simulator that runs at up to 16 FPS on eight RTX 5090 GPUs, showing that camera control, long-horizon memory, and few-step inference can coexist without a giant model (5B params beating 8B and 14B baselines on their own benchmark). For practitioners it mostly matters as a reference architecture, since the practical result depends on an 8-GPU cluster and evaluation on the authors' own suite.
Practical takeaway
Watch the linked GitHub repo if you want the E-PRoPE camera-conditioning trick, which claims comparable trajectory-following to full PRoPE while cutting inference latency about 30% by running projective attention on 4.5x downsampled tokens. Also note their revisit-consistency memory benchmark, a more useful way to test whether a world model actually remembers scenes than FVD/FID.
Key result
On the authors' own 5-second basic evaluation, DreamX-World-1.0-5B scored 73.75 camera control and 84.76 overall, versus 80.79 (HY-WorldPlay 1.5, 8B) and 80.45 (LingBot-World, 14B); note this is a self-devised benchmark and trajectory set, not an external standard, and the gap narrows on 30-second rollouts (70.41 vs 68.85 and 67.43).
Subscribe
Get the next issue.
Free. One email a week. Unsubscribe any time: no account, no dark patterns.