Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,34 @@
+---
+license: mit
+tags:
+  - 3d-editing
+  - trellis
+  - controlnet
+---
+
+# PartFlow — model weights
+
+Trained ControlNet denoisers for **PartFlow**, a two-stage image-conditioned
+3D editing model built on [TRELLIS](https://github.com/microsoft/TRELLIS).
+
+Inference code: see the PartFlow GitHub repository.
+
+## Contents
+
+| Path | Model | Description |
+|---|---|---|
+| `stage1_ss/denoiser_ema0.9999_step0080000.pt` | `ControlledSparseStructureFlowModel` | Stage 1 — sparse-structure ControlNet (EMA, step 80k) |
+| `stage1_ss/config.json` | — | Stage 1 model config |
+| `stage2_slat/denoiser_ema0.9999_step0080000.pt` | `ControlledSLatFlowModel` | Stage 2 — structured-latent ControlNet (EMA, step 80k) |
+| `stage2_slat/config.json` | — | Stage 2 model config |
+
+These are EMA weights (decay 0.9999). The frozen DINOv2 encoder and TRELLIS
+SS/SLat decoders are not included — PartFlow fetches them from their official
+repos at inference time.
+
+## Usage
+
+```bash
+python download_weights.py        # pulls this repo into ./weights/
+python inference.py --input examples/add_ring_band --output_dir outputs
+```