---
license: other
license_name: per-model
pipeline_tag: image-to-3d
tags: [onnx, image-to-3d, depth-estimation, triposr, modelado]
---

# Modelado models (ONNX)

The ONNX models [Modelado](https://github.com/alegauss/modelado) downloads to turn a photo
into a 3D model on Windows, with DirectML or on the CPU. They are converted from each
upstream's own weights and checked against PyTorch by `scripts/export_models.py`. Every
file is listed in `manifest.json` with its size and SHA-256, and the app refuses one that
differs. Each model keeps its upstream license:

| Model | Task | Variants | License |
| --- | --- | --- | --- |
| triposr-encoder | image-to-triplane | fp32 (1675 MB), fp16 (839 MB) | [MIT](https://huggingface.co/stabilityai/TripoSR/blob/5b521936b01fbe1890f6f9baed0254ab6351c04a/README.md) |
| triposr-decoder | triplane-decoder | fp32 (0.2 MB) | [MIT](https://huggingface.co/stabilityai/TripoSR/blob/5b521936b01fbe1890f6f9baed0254ab6351c04a/README.md) |
| depth-anything-v2-small | depth | fp32 (99 MB), fp16 (50 MB) | [Apache-2.0](https://huggingface.co/depth-anything/Depth-Anything-V2-Small-hf/blob/5426e4f0f36572d16453bbda7a8389317b1bef99/README.md) |
| u2net | background-removal | fp32 (176 MB) | [Apache-2.0 (U-2-Net weights), MIT (rembg export)](https://github.com/xuebinqin/U-2-Net/blob/master/LICENSE) |
| realesrgan-x4plus | upscale | fp32 (67 MB), fp16 (34 MB) | [BSD-3-Clause](https://github.com/xinntao/Real-ESRGAN/blob/master/LICENSE) |

`triposr-encoder` and `triposr-decoder` are [TripoSR](https://huggingface.co/stabilityai/TripoSR)
split in two: the encoder turns a 512 x 512 photo of an object into a triplane, the decoder
returns density and colour at any point of it, and the app extracts the mesh with marching cubes.