Hy-Nomas

Hy-Nomas

Generates a video world one segment at a time from text instructions and camera moves, so you can change objects, style and viewpoint as it goes. For creators and researchers.

Made by tencent on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Hy-Nomas is a Tencent model that makes video worlds you can steer over several turns. You use text instructions and camera control. It builds the world one video segment at a time, so you can add objects, change a scene or move the camera while generation goes on.

Who it suits: It suits creators and researchers who want to direct a generated video world step by step instead of getting one fixed clip.

Industry | Who uses it | How they use it | Example
Film and video | Creators | Build a scene step by step from text | A world where a new character and a style change arrive in turn
Games | Concept artists | Explore a made-up space with camera control | A walk through an invented environment
Research | AI researchers | Test interactive control and consistency | A run on the Hy-Nomas-Bench cases

People who need one quick clip from a single prompt can skip it, since the model is built for multi-turn worlds made segment by segment.

Where to use it:

• A web demo, which the card says you can use to interact with the model
• The standard sequence-parallel setup, or the single-GPU setup for an NVIDIA GeForce RTX 4090
• Windows, macOS or Linux, where you download the files with Python

Use case | Who | What they get back
Text-guided segments | Creators | The next video segment, with the objects, style or action you describe
Camera-controlled shots | Filmmakers | Viewpoint and movement that follow a camera path you supply
Consistency tests | Researchers | Four benchmarks of 280 cases on control and spatial memory

How to set it up: You need these first:

• Python and the huggingface_hub package
• A GPU setup, either a multi-GPU sequence-parallel one or a single RTX 4090 with CPU offload
• The run steps from the card's code repository, which the card points to

1. Install the Hugging Face hub package.
2. Download the model files to your computer.
3. Follow the card's run steps for your GPU setup.

pip install -U huggingface_hub
hf download tencent/Hy-Nomas

How to use it: Write a text instruction for the first segment, and add a camera trajectory when you want to control the viewpoint. Then give a new instruction for each next segment. A cache carries the earlier context along, so the same world continues.

Example: You start with an instruction that describes a scene. For the next segment you ask for a new character and a slow camera move to the left. The card says each new segment is built from your text, the camera motion and the cached history.

For newcomers and experts: A world model makes video that you can keep changing as it plays. A first small thing to try is one scene instruction and a single follow-up that changes the style.

For experts, each segment is denoised inside a compressed video latent space, and the segments follow one another autoregressively. It combines a diffusion transformer, a Qwen text encoder, camera modules and a PixelUMM decoder, with 54 dual-stream blocks. The distilled setup uses 8 denoising steps for 720 by 1280 output at 24 FPS. The single-GPU setup trades precision for memory with INT8 weights, an INT8 cache, an FP8 decoder and CPU offload.

Write a text instruction
Add a camera path if you want one
The model generates one video segment
A cache carries the context forward
Write the next instruction

Good to know:

• Output: 720 by 1280 at 24 FPS in the distilled setup
• Benchmark: Hy-Nomas-Bench has four benchmarks with 280 test cases
• Variants: Hugging Face lists 3 quantized and 2 fine-tuned versions made by other people
• License: apache-2.0
• Cost to run: none named, no API key or account. The web demo is a separate service
• Data it reaches: it downloads model files from Hugging Face and works on the text and camera paths you give it
• Not for: none named in the card text shown

On Hugging Face, 13 November 2026

• Library: diffusion-single-file
• Formats: Safetensors
• Downloads in the last 30 days: 90,896
• Likes: 153
• License: apache-2.0
• Quantized versions: 3
• Fine-tunes: 2
• Tags: diffusion-single-file, diffusers, safetensors, hy-nomas, world-model, video-generation, interactive-video-generation, camera-control, autoregressive, diffusion-transformer, pytorch

Links

• Model: https://huggingface.co/tencent/Hy-Nomas
• Files: https://huggingface.co/tencent/Hy-Nomas/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download tencent/Hy-Nomas

Then follow the usage in its model card. Nothing in this kit runs by itself.