# Image to 3D game asset (TRELLIS.2 in GGUF) > Open image-to-3D model: one picture in, a textured 3D mesh out. TRELLIS.2-4B > packed as a single GGUF file that runs locally on the diffusion engine of the > `ggk` or `gguf-core` Python package; `--serve` puts it behind an OpenAI-style > endpoint, `POST /v1/images/edits`, which takes the image and answers with the > mesh file (.glb, .ply or .obj). Quick start: ```bash GGK_CUDA=1 pip install ggk # or: GGK_CUDA=1 pip install gguf-core (command `gguf` instead of `ggk`) ggk -m trellis2-4b-nvfp4.gguf --serve --port 8888 curl -f http://127.0.0.1:8888/v1/images/edits -F image=@sheep.png --output sheep.glb ``` The response arrives when the mesh is done: about 80 seconds on a 6 GB GPU, minutes for the high-resolution pipelines. Wait for it; do not retry early. ## Docs - [Complete guide](llms-full.txt): install, the model file, server options, every request field, output formats, timings, pitfalls. Self-contained; read this one. - [Website](index.html): the same documentation for people, plus a browser playground with a 3D viewer. ## Sources - [Model files](https://huggingface.co/gguf-org/trellis2-4b-gguf): the TRELLIS.2 pipeline as one GGUF - [TRELLIS.2](https://github.com/microsoft/TRELLIS.2): the original model - [ggk](https://github.com/gguf-io/gk): the runtime - [gguf-core](https://github.com/calcuis/gguf-core): the same runtime under the `gguf` command