SSyncDev

Helix World

SyncDev model directory · Profile updated Nov 17, 2026

Compare with another model

Helix World (preview) is an open real-time world model from Noiz AI and HKUST that streams explorable 768x512 video with stereo sound that follows the camera.

Key specifications

Specification score
14.8
License
Open weights

The specification score summarizes recorded features and limits. It is not a benchmark result.

Inputs and outputs

Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.

Accepts

  • Text input
  • Image input

Produces

  • Video output
  • Audio output

Capabilities

VisionAudioOpen weights

What is Helix World?

Interactive world models have mostly been silent. Helix World (preview), from Noiz AI and the Hong Kong University of Science and Technology, adds the missing channel: as you steer the camera through a generated scene, stereo sound turns with it. Its public launch is dated 12 October 2026; the accompanying paper is dated 29 September.

How a session works

You start from one image, either an example or your own upload, plus an optional prompt. Leave the prompt empty and the scene is described from the frame automatically. WASD moves the camera and the arrow keys turn it. Video and audio arrive in chunks of about 1.3 seconds, so a new key press or prompt takes effect from the next chunk, not instantly.

Under the hood it extends the open LTX-2.3 audio-visual model with continuous 6-DoF camera control and keyboard actions. A bidirectional teacher was distilled into a few-step causal student that streams with a sliding key-value cache.

Specifications

Item Detail
Developers Noiz AI and HKUST
Base LTX-2.3
Video 768x512, 24 fps
Audio 48 kHz stereo, spatially consistent with the camera
Chunk length ~1.3 s
Real-time factor 0.77 on one H800, decoding included
Inputs Image, optional text, camera and keyboard actions
Weights Preview checkpoint, NoizAI/HelixWorld-preview
API None

A real-time factor below 1.0 means it renders faster than playback. That figure is for a datacentre GPU; no consumer-GPU numbers are published, so do not assume it will keep up on a workstation card.

The team reports a WBench navigation average of 79.9 and a spatial-acoustic consistency score of 41.8 on its own HelixBench. It is candid that the best silent models still edge it on visuals alone, 79.9 against 82.0, and that sound onset timing needs work.

Licence: check before you build

Here the sources do not line up. The launch article states no licence. The repository's metadata lists Apache-2.0, while its description calls the release a private inference checkpoint and points to a separate licence file for the full terms. Read that file, and if you plan anything commercial, get the terms in writing.

What it is good for today

Research and prototyping. Game teams exploring generated environments, audio researchers studying spatial sound, and anyone building embodied-agent simulations can get real value from a streaming model that does picture and sound together. Production use is premature: the output is capped at 768x512 and this is explicitly a preview.

Kicking the tyres

Try the web demo first; it costs nothing. Then run the checkpoint on your own GPU with a handful of source images you care about, measure real-time factor on that hardware, and walk the camera in circles. Two failure modes are worth watching: scene drift on long walks and audio that lags a turn.

For video models without interactivity, browse image-to-video or image-text-to-video.

How to evaluate Helix World

Compare the model on the work you need to complete.

Tasks and inputs

Helix World is listed for Image-text-to-video, Image-to-video. It accepts text, image and produces video, audio. Use the publisher’s input format and preprocessing for your chosen deployment.

Capacity and cost

Choose duration, resolution and frame rate together. Compare the cost of an accepted clip, including repeat generations.

Choosing a deployment

The weights are downloadable under the publisher’s model licence. Check runtime support and size the complete model and working context for your expected concurrency.

Frequently asked questions

No. Only a preview checkpoint, code and a web demo have been released.

These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.

tbc

TBC mischief 1

TBC mischief 1 is an image-to-video model built on Wan 2.2, sold on fal and OpenRouter at about half a cent per second for 3-5 s clips at up to 720p.

— context

NVIDIA

Cosmos3-H-Dreams

Cosmos3-H-Dreams is NVIDIA's open real-time surgical world model: it streams Versius robot video from one endoscope frame and two-arm commands.

— context

NVIDIA

Cosmos3-H-Embodied-Nano

Cosmos3-H-Embodied-Nano is NVIDIA's open 16B surgical world model that does forward dynamics, inverse dynamics and policy from one checkpoint.

— context

NVIDIA

Cosmos3-Nano-Sim-Bimanual

Cosmos3-Nano-Sim-Bimanual is NVIDIA's open 16B world model that simulates video of a two-armed AgiBot robot as it moves its camera and manipulates.

— context

Alibaba

Wan-Animate-2

Wan-Animate-2 is Alibaba Wan's open 14B character-animation model that drives a character from video directly, with text-controlled camera angle.

— context

Z.ai / Zhipu AI

SCAIL-2

SCAIL-2 is Zhipu's open character-animation model that drives a reference character from any video, with no pose skeletons or masks. MIT licence.

— context