
Picture published by the creator on Hugging Face.
Qwen3.5-4B-3D-Coder: Turn prompts into 3D Three.js web scenes (4B)
Added · Updated
Made by Caio Giovannetti
View profile on Hugging Face (opens in a new tab)Describe a scene or attach a picture and get a single web page that draws it in interactive 3D.
For developers who build 3D scenes for the web.
Model details
Original model ccore/Qwen3.5-4B-3D-Coder
Input Unknown Output Unknown
Ways to use it Run locally
Role in this family: Fine-tuned version.
Qwen/Qwen3.5-4B (opens in a new tab)What it does
Qwen3.5-4B-3D-Coder is an independent retrained version of Qwen3.5-4B, trained on 4,146 example scenes. You give it a system prompt asking for the Three.js 3D library and a scene description; it writes one self-contained HTML and JavaScript file that a web browser opens as an interactive 3D scene. With the image add-on file loaded, it can also work from a reference image.
How to use it
Load the weights in a runtime: download a GGUF file (a compressed model format) and run it with llama.cpp's server, then open its web page, attach an image if you like, and send a prompt. In LM Studio, turn off multi-token prediction first. Full FP16 weights also load with Transformers. Use a randomness setting (temperature) of 0.6–0.8 and the source's Three.js system prompt.
Common use cases
- Ask for one self-contained HTML file showing a canal lock and weir complex with a central island, and you get a web page rendering that 3D scene.
- Attach a reference picture and type 'create this scene using voxels 3d', and you get a Three.js scene built from blocks that interprets the picture, not every detail.
- Describe an air traffic control tower with rotating radar and light signals, and you get a standalone HTML page meant to animate the tower, radar and beacon.
Can I use this?
Requirements Download a GGUF file and, for image input, the projector file once; the same projector works with either GGUF. Apply the generation settings listed in the model card. Licensed under Apache-2.0.
Caveat Image tests were short, not full-context stress tests, and image support in other apps depends on how those apps handle images.
Worth knowing Early preview still in training; some scenes need another attempt or a correction. Image input works only with the GGUF versions plus the projector. Not an official Qwen release.
Why it's here
- 904 people have liked it on Hugging Face.
- It was downloaded 254,053 times in the last 30 days.
Numbers as of 8 November 2026, from BuildTube’s latest daily refresh.
Behind it
View model pageBuildTube has not run or verified this project. Everything above is written from what the creator published.
Made this?
Quick questions
- What is Qwen3.5-4B-3D-Coder?
- Qwen3.5-4B-3D-Coder is an independent retrained version of Qwen3.5-4B, trained on 4,146 example scenes.
- What can I use Qwen3.5-4B-3D-Coder for?
- Ask for one self-contained HTML file showing a canal lock and weir complex with a central island, and you get a web page rendering that 3D scene; Attach a reference picture and type 'create this scene using voxels 3d', and you get a Three.js scene built from blocks that interprets the picture, not every detail; Describe an air traffic control tower with rotating radar and light signals, and you get a standalone HTML page meant to animate the tower, radar and beacon
- Can I run Qwen3.5-4B-3D-Coder on my computer?
- It runs on your own hardware with a model runtime. The model card mentions GGUF, llama.cpp, Transformers; check its setup and hardware notes.
- What do I need to use Qwen3.5-4B-3D-Coder?
- A runtime such as llama.cpp, LM Studio or Transformers. Image input needs the separate projector file of about 644 MiB. Memory and GPU needs are not documented.
- What license does Qwen3.5-4B-3D-Coder use?
- Licensed under Apache-2.0.
More from this week
- Idol Turn one picture of an object into a 3D Gaussian splat
Idol turns one photo, render or generated image of an object into a 3D Gaussian splat file.
- P-H3-Image-Comfy Make and edit 8-megapixel images in ComfyUI
Turn short text requests into 8-megapixel images and edit pictures on your own graphics card.
- agent-desktop Let AI agents read and click desktop apps
For developers building computer-use agents: agent-desktop lets an AI agent read and operate desktop apps on Mac, Windows or Linux from a command line.
Explore by topic
Post this
Download share imageFor makers
Made this? Show it’s on BuildTube — add the badge to your README or embed the card on your site.
README badge (Markdown)
[](https://buildtube.app/p/hf-ccore-qwen3-5-4b-3d-coder/)Embed card (HTML)
<iframe src="https://buildtube.app/embed/hf-ccore-qwen3-5-4b-3d-coder/" width="560" height="240" style="border:0;max-width:100%" loading="lazy" title="Qwen3.5-4B-3D-Coder on BuildTube"></iframe>