Add optional vision projector and image-guided tree example

Add the original Qwen3.5-4B BF16 vision projector, tree reference and generated scene screenshots, and llama.cpp usage instructions. Preserve all existing model weights and other release files.

.gitattributes

@@ -49,3 +49,6 @@ images/dif.png filter=lfs diff=lfs merge=lfs -text
 images/roma3.png filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
 Qwen3.5-4B-3D-Coder_Technical_Report.pdf filter=lfs diff=lfs merge=lfs -text
+vision-tree-reference.png filter=lfs diff=lfs merge=lfs -text
+vision-tree-example.png filter=lfs diff=lfs merge=lfs -text
+mmproj-Qwen3.5-4B-BF16.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -22,6 +22,48 @@ This is an early preview of a 4B model I am still training, with a particular fo
 
 This is an **independent fine-tune** by [CCORE](https://huggingface.co/ccore), based on [Qwen/Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B). **It is not an official Qwen release.**
 
+## Optional vision support — October 30, 2026
+
+The **Q4_K_M and Q8_0 GGUF versions can also accept image inputs** when loaded with the original Qwen3.5-4B vision encoder/projector. You can attach a reference image and ask the model to generate a Three.js scene from it.
+
+[**Download the optional BF16 vision projector**](https://huggingface.co/ccore/Qwen3.5-4B-3D-Coder/resolve/main/mmproj-Qwen3.5-4B-BF16.gguf?download=true) — approximately **644 MiB**. Download it once; the same file is used with either GGUF quantization.
+
+### Image-guided Three.js example
+
+With a reference image attached and the Three.js system prompt below, the user prompt was simply:
+
+```text
+create this scene using voxels 3d
+```
+
+<table>
+<thead><tr><th>Input reference</th><th>Generated Three.js scene</th></tr></thead>
+<tbody><tr>
+<td width="40%" align="center"><a href="https://huggingface.co/ccore/Qwen3.5-4B-3D-Coder/blob/main/vision-tree-reference.png"><img src="vision-tree-reference.png" alt="Input reference: an orange voxel tree on a floating island" width="404" /></a></td>
+<td width="60%" align="center"><a href="https://huggingface.co/ccore/Qwen3.5-4B-3D-Coder/blob/main/vision-tree-example.png"><img src="vision-tree-example.png" alt="Generated Three.js scene: Tree Island" width="1000" /></a></td>
+</tr></tbody>
+</table>
+
+The reference image was supplied in [community discussion #9](https://huggingface.co/ccore/Qwen3.5-4B-3D-Coder/discussions/9); credit for the reference belongs to its original creator. The right-hand image is a screenshot of the generated scene. This is a qualitative example of image-guided generation; the output interprets the reference and does not reproduce every detail.
+
+### Load the GGUF with vision in llama.cpp
+
+Download a model GGUF and the projector, then load both files explicitly:
+
+```bash
+llama-server \
+  -m Qwen3.5-4B-3D-Coder-Q4_K_M.gguf \
+  --mmproj mmproj-Qwen3.5-4B-BF16.gguf \
+  --n-gpu-layers all --flash-attn on \
+  --ctx-size 65536 --parallel 1
+```
+
+Open the server's WebUI, attach an image, and send your prompt. The setup was tested locally with **llama.cpp b11889**. For Q8_0, change the model filename to `Qwen3.5-4B-3D-Coder-Q8_0.gguf`; `--cache-type-k q8_0 --cache-type-v q8_0` can reduce KV-cache memory use and was tested locally with vision at a configured 64K context. These were short tests, not full-context stress tests.
+
+The projector uses additional memory when loaded, and image tokens share the context window with the prompt and generated code. Image support in other frontends depends on their multimodal implementation.
+
+**Projector provenance:** this is the unmodified BF16 projector from [lmstudio-community/Qwen3.5-4B-GGUF](https://huggingface.co/lmstudio-community/Qwen3.5-4B-GGUF/tree/f9f88ac3e234be915e23811a6d28ea287bdb927e), based on [Qwen/Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B), released under Apache 2.0. It is an optional companion file for the existing GGUF weights.
+
 ## Technical report — now available
 
 The **technical report for this training experiment is now available**.

mmproj-Qwen3.5-4B-BF16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae08d9d7eceb8f2d0672d61b5e6aa78b611f2942b55ed71d21414980cc454b91
+size 675568768