CCORE GLB-3D-Coder 1.7B

CCORE GLB-3D-Coder 1.7B

Writes a complete Three.js page for a game asset you describe, with a 3D preview and an Export GLB button. A fine-tuned Qwen3-1.7B in GGUF that runs on modest hardware, for game developers.

Made by ccore 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: This is a Qwen3-1.7B model fine-tuned on a GLB asset dataset. You describe a game asset in plain words, and it writes one standalone HTML page that shows the asset in 3D and includes a button to export it as a GLB file. The author calls this an initial test checkpoint.

Who it suits: It suits game developers, 3D hobbyists and prototypers who have limited VRAM or no dedicated GPU and want a quick loop for low-poly props and creatures.

Title: Who might use it
Industry | Who uses it | How they use it | Example
Game development | Indie developers | Generate low-poly props and enemies for a Godot project | A stone golem boss with a glowing crystal core
Prototyping | Designers | Draft placeholder 3D assets quickly | A bakery counter as a placeholder in a shop scene
Education | Hobbyists | Read the generated Three.js pages to learn how a scene is built | A low-poly iron table
Technical art | 3D artists | Pull the exported GLB into an editor and refine it | A magic sword on a display pedestal

People who need final, polished assets or a finished release can skip it for now, since the weights are still being updated.

Where to use it:

• llama.cpp: a Qwen3-compatible runtime that loads GGUF files
• LM Studio: another runtime the card names
• A web browser: opens the generated HTML page and its 3D preview
• A 3D editor or the Godot editor: opens the exported GLB file
• Windows, macOS or Linux: download the files with Python

Title: Jobs it does
Use case | Who | What they get back
Asset generation | Game developers | A standalone HTML page that shows a low-poly asset
GLB export | 3D artists | A GLB file downloaded from the page's Export GLB button
Quick iteration | Hobbyists | A short loop on a CPU or a small GPU

How to set it up: You need these first:

• Python on Windows, macOS or Linux, to download the files
• A runtime that supports Qwen3 GGUF files, such as llama.cpp or LM Studio
• Room for the file you choose, about 1.83 GB for Q8_0 or 3.45 GB for F16

1. Install the Hugging Face hub package.
2. Download the model files.
3. Load a GGUF file in your runtime. The card suggests Q8_0 as a starting point.

pip install -U huggingface_hub
hf download ccore/Qwen3-1.7B-GLB-3D-Coder

How to use it: Load a GGUF file, give it the system prompt shown here, then ask for the asset in your own words. Save the HTML it writes, open it in a browser and press the export button.

Create game assets for the Godot editor, which uses GLB files. Create whatever the user requests as a good-quality low-poly asset using only Three.js in one complete standalone HTML file. The HTML must show the asset and include an Export GLB button that downloads a working GLB. If the requested asset needs animation, add it and include the animation in the exported GLB; otherwise make it static.

The card recommends these sampling values.

Title: Recommended sampling
Setting | Value
Temperature | 0.8
Top K | 40
Top P | 0.95
Min P | 0.0
Repeat penalty | 1.0

Example: You ask for a stone golem boss with chunky rock arms and legs and a glowing green crystal at its core. The model writes one HTML page that shows the golem in 3D with an Export GLB button, and you save the page and open it.

For newcomers and experts: If you are new to this, load the Q8_0 file in LM Studio, paste the system prompt and ask for a simple low-poly iron table. Open the saved page in a browser and look at the result.

If you work in the field, the base is Qwen3-1.7B with full causal attention and grouped-query attention across all 28 layers. The author chose a small model for a short loop on limited hardware. Generation speed varies with your machine, the runtime, the context size and the amount of code produced. Safetensors and fuller training details are promised after training ends.

Describe an asset in plain words
The model writes one HTML page
Open the page to preview in 3D
Click Export GLB
Open the GLB in your 3D editor

Good to know: The card lists two GGUF files and their approximate sizes.

Title: GGUF file size in GB
Q8_0: 1.83 GB
F16: 3.45 GB

• License: apache-2.0
• Cost to run: none named
• Data it reaches: the prompts you give it and the HTML files you save and open yourself
• Not for: a final release, since the author calls this an initial test checkpoint and plans updated weights

On Hugging Face, 13 November 2026

• Task: text generation
• Library: gguf
• Parameters: 1.7B
• Weights: about 860 MB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 421
• Likes: 13
• License: apache-2.0
• Based on: Qwen/Qwen3-1.7B
• Tags: gguf, code, threejs, 3d, godot, text-generation, conversational

Links

• Model: https://huggingface.co/ccore/Qwen3-1.7B-GLB-3D-Coder
• Files: https://huggingface.co/ccore/Qwen3-1.7B-GLB-3D-Coder/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download ccore/Qwen3-1.7B-GLB-3D-Coder

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