Open-source interactive real-time world generator with memory for long-term consistency. Built on Alibaba's Wan 2.2, with a 4-step turbo model; code released.
WorldPlay 2
Real-time interactive world model with compressed memory and hybrid action and semantic control, built on Wan 2.2.
Updated EU status checked Last shown
WorldPlay 2 at a glance
- What is WorldPlay 2?
- WorldPlay 2 is a model. Real-time interactive world model with compressed memory and hybrid action and semantic control, built on Wan 2.2.
- Does WorldPlay 2 work in the EU?
- Unverified. No explicit evidence found. Last verified 14 Nov 2026.
- How much does WorldPlay 2 cost?
- Open source. Code reported as released (per a YouTube creator); no pricing found on the project page.
- Where was WorldPlay 2 shown?
- In 1 creator video by AI Search. Most recently by AI Search on 11 Oct 2026, at 1:06 in “New AI enzymes, 700+ math breakthroughs, AI for DNA mutations, room temp semiconductors: AI NEWS”.
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
WorldPlay2 is a research interactive world model that generates video of a scene in real time in response to user controls, while keeping the world consistent over long horizons. According to a YouTube review, it is built on Alibaba's Wan 2.2, includes a 4-step turbo model, and has its code released. The project page is titled "Extending Real-Time Interactive World Models in Control and Horizon".
It uses a factorized hybrid control interface. Frame-aligned action control covers camera pitch and yaw, discrete forward/backward and lateral movement, camera perspective and jump. Structured semantic control splits a caption into scene, character and event fields, so appearance, identity and dynamic events (for example igniting a stove) can be controlled separately.
For long-horizon consistency, a learnable history compressor turns past context into compact memory tokens (sink, compressed and adjacent temporal tokens), cutting sequence length by about 32x. A distillation method called Stable Forcing (few-step initialization plus full-rollout replay) trains an autoregressive few-step student from a bidirectional teacher. The page shows demos and comparisons with other methods; no pricing, license, repository or paper links were present in the provided material.
- Real-time interactive world generation
- Frame-aligned action control (camera, movement, jump)
- Structured semantic control: scene, character, event
- Compressed memory tokens (~32x shorter context) for long-horizon consistency
- Stable Forcing distillation to a few-step student (4-step turbo model reported)
- Built on Wan 2.2 (per creator report)
As seen on YouTube
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Similar AI tools
Closest neighbours by meaning, not just by category.
GWM Worlds 2
Model · Runway
Runway's real-time interactive world model generating 720p/24fps video with audio, steered by text actions and camera input.
Alaya World
Model
Interactive autoregressive world model generating minute-long playable video with real-time camera control and prompt events.
SolarWM
Model
Open framework and dataset for turning video diffusion models into real-time, long-horizon interactive world models.
Streaming Code World Model
Model
Research code that streams video generation chunk by chunk from a rough 3D proxy, built on Code World Model.
H3 World
Model
Research framework that turns the Minimax H3 video generator into an interactive world model controlled by key presses.
Evoke
Model
14B autoregressive world model that generates interactive, camera-controlled video worlds in 3 denoising steps.