Hy-Nomas Tencent Studio

Hy-Nomas Tencent Studio

Lets you steer a generated world turn by turn with Say and Move commands, camera steering and MP4 export, in a Tencent-styled studio around the Hy-Nomas model. For people exploring interactive world generation.

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

About

What it is: Hy-Nomas Tencent Studio is a redesigned front end for the Hy-Nomas world-generation demo. You compose a world, then steer it turn by turn with Say and Move commands and camera steering. Each turn lands as a clip you can export as MP4. It runs on ZeroGPU, and you can open it in the browser.

Who it suits: It suits people who want to try interactive world generation, and developers who want to see how a custom front end is laid over an existing demo backend.

Industry | Who uses it | How they use it | Example
Gaming | Game designers | Try ideas for generated worlds | Pick a gallery preset and steer it
Film and video | Creators | Build short clips from a generated world | Export the timeline as an MP4
Research | AI researchers | Explore interactive world generation | Use Say and Move turns to see how the world responds
Web development | Front-end developers | Study a custom studio-style interface | Read index.html and app.py

Skip it if you want a language model to rewrite your prompts, since this version removed that step.

Where to use it:

• A web browser, on the Space's page at Hugging Face, which runs on ZeroGPU
• Your own Windows, macOS or Linux computer with Python 3.12

Use case | Who | What they get back
Generate a world | Creators | A live world that you steer turn by turn
Steer with the camera | Designers | Camera movement as part of each turn
Export the result | Video makers | An MP4 built from the clips on the timeline

How to set it up: The online page needs nothing. A local copy needs Python 3.12 and the packages in requirements.txt. Live sessions need ZeroGPU on the Space, and no secrets are required.

• Python 3.12 and pip
• Python packages from requirements.txt
• ZeroGPU enabled on the Space, for live sessions
• Windows, macOS or Linux

1. Install the Hugging Face hub library.
2. Download the Space's files into a folder named hy-nomas.
3. Move into that folder.
4. Install Gradio 6.30.0, then the packages from requirements.txt.
5. Start app.py.

pip install -U huggingface_hub
hf download tencent/hy-nomas --repo-type space --local-dir hy-nomas
cd hy-nomas
pip install gradio==6.30.0
pip install -r requirements.txt
python app.py

How to use it: Open the Space's page, or run the last command above for a local copy and open the address it prints. The workspace has four areas.

Tool | What it does
Project panel | On the left, holds the world composer and the gallery presets
Program Monitor | In the center, shows the world with the director's Say and Move panel
Inspector | On the right, holds the director's notes
Timeline | Runs along the bottom. Every turn becomes a clip there, with thumbnails, playback controls and an MP4 export option

Example: You pick a gallery preset in the project panel and give a Say turn to describe what happens next. A new clip appears on the timeline, and you step through it with the transport controls.

For newcomers and experts: Start from a gallery preset and give one short Say turn. Watch the clip appear before you try Move or camera steering.

When the Space starts, app.py pulls the upstream demo, which holds the session backend, the pipeline code and the gallery assets, and lays this repository's index.html over it. The language model prompt enhancer is gone. Each turn is planned by two built-in deterministic builders that emit the structured segment format the model was trained on. ZeroGPU session leasing, undo and export behave as in the original. The design is dark, with an emerald accent for selection and primary actions.

The Space pulls the upstream demo at startup
Its index.html is laid over the original
Your Say or Move turn is planned locally
The model renders the turn on a ZeroGPU
The clip lands on the timeline

Good to know:

• Change from the original: The language model prompt enhancer is removed, so no token or secret is needed.
• Needs: ZeroGPU must be enabled for live sessions.
• License: apache-2. The upstream code is also Apache 2.0.
• Cost to run: none named. No secrets are needed.
• Data it reaches: The upstream demo code and assets it pulls at startup, and the worlds and turns you enter.

On Hugging Face, 13 November 2026

• Likes: 26
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/tencent/hy-nomas
• Files: https://huggingface.co/spaces/tencent/hy-nomas/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.