Upload README.md with huggingface_hub

README.md

@@ -1,10 +1,16 @@
 ---
-license: cc0-1.0
+license: other
+license_name: per-model
+license_link: https://github.com/fernandotonon/QtMeshEditor/blob/master/THIRD_PARTY_AI_MODELS.md
 tags:
 - onnx
+- gguf
 - pbr
 - texture
 - normal-map
+- 3d
+- rigging
+- image-to-3d
 - qtmesheditor
 - qtmesh
 - qtmesh-cloud
@@ -13,69 +19,99 @@ library_name: onnx
 
 # QtMeshEditor — AI models
 
-ONNX models used by [QtMeshEditor](https://github.com/fernandotonon/QtMeshEditor)'s
-AI-assisted authoring features.
-
-## PBR map synthesis 
+The models used by [QtMeshEditor](https://github.com/fernandotonon/QtMeshEditor)'s
+AI-assisted authoring features. **This repo is what the app downloads from at
+runtime** (each model on first use, then it runs locally/offline).
+
+**Licenses are per model** — see the table and each dedicated repo. The
+dedicated repos carry the full model cards (I/O contracts, provenance,
+reproduction scripts) for anyone who wants the converted weights standalone.
+
+| folder / files | feature | dedicated repo (full card) | license |
+|---|---|---|---|
+| `1x-PBRify_*.onnx` | PBR maps from albedo | [QtMeshEditor-pbrify-onnx](https://huggingface.co/fernandotonon/QtMeshEditor-pbrify-onnx) | CC0-1.0 |
+| `RealESRGAN_x{2,4}plus.onnx` | texture upscaling | — (straight Real-ESRGAN export) | BSD-3-Clause |
+| `unirig/` | auto-rig skeleton prediction | [QtMeshEditor-unirig-onnx](https://huggingface.co/fernandotonon/QtMeshEditor-unirig-onnx) | MIT |
+| `skintokens/` | ML skin-weight prediction | [QtMeshEditor-skintokens-onnx](https://huggingface.co/fernandotonon/QtMeshEditor-skintokens-onnx) | MIT |
+| `triposr/` | image → 3D (triplane) | [QtMeshEditor-triposr-onnx](https://huggingface.co/fernandotonon/QtMeshEditor-triposr-onnx) | MIT |
+| `triposg/` | image → 3D (rectified-flow DiT) | [QtMeshEditor-triposg-onnx](https://huggingface.co/fernandotonon/QtMeshEditor-triposg-onnx) | MIT |
+| `inbetween/rmib.onnx` | animation in-betweening (ours) | [QtMeshEditor-rmib-inbetween](https://huggingface.co/fernandotonon/QtMeshEditor-rmib-inbetween) | CC-BY-4.0 |
+| `motion/` | text-to-motion + clip library (ours) | [QtMeshEditor-t2m](https://huggingface.co/fernandotonon/QtMeshEditor-t2m) | CC0-1.0 |
+| `segment/meshseg.onnx` | mesh part segmentation (ours) | [QtMeshEditor-mesh-segmentation](https://huggingface.co/fernandotonon/QtMeshEditor-mesh-segmentation) | CC-BY-4.0 |
+| `rembg/u2net.onnx` | background removal | — (standard U²-Net) | Apache-2.0 |
+| `caption/SmolVLM-500M-*.gguf` | image captioning | — (quantized SmolVLM) | Apache-2.0 |
+
+## PBR map synthesis
 
 `1x-PBRify_NormalV3.onnx`, `1x-PBRify_RoughnessV2.onnx`, `1x-PBRify_Height.onnx`
 generate tangent-space normal / roughness / height maps from a single albedo
-(diffuse) texture.
-
-These are **ONNX re-exports** of the CC0 SPAN models from
-**[Kim2091/PBRify_Remix](https://github.com/Kim2091/PBRify_Remix)** (LICENSE:
-CC0-1.0), trained on CC0 content from ambientCG / Poly Haven. Converted with
-`scripts/export-pbrify-onnx.py` in the QtMeshEditor repo (spandrel +
-`torch.onnx.export`, opset 18). All credit for the weights goes to Kim2091.
-
-- **License:** CC0-1.0 (public domain), same as the source models.
-- **I/O:** 1×3×H×W float NCHW in `[0,1]` → 1×3×H×W out (normal as RGB;
-  roughness/height as RGB, consumed as luminance). Dynamic H/W.
+texture. ONNX re-exports of the CC0 SPAN models from
+**[Kim2091/PBRify_Remix](https://github.com/Kim2091/PBRify_Remix)** — all
+credit to Kim2091. I/O: `1×3×H×W` float `[0,1]` → `1×3×H×W`, dynamic H/W.
 
-QtMeshEditor downloads these on first use into `<AppData>/ai_models/pbr/`.
-
-More info in [QtMesh Cloud website](https://qtmesh.dev)
 ## Texture upscaling
 
 `RealESRGAN_x2plus.onnx`, `RealESRGAN_x4plus.onnx` — 2×/4× super-resolution.
-ONNX re-exports of **Real-ESRGAN** ([xinntao](https://github.com/xinntao/Real-ESRGAN),
-**BSD-3-Clause**). Downloaded into `<AppData>/ai_models/pbr/`. Credit: xinntao.
+ONNX re-exports of **Real-ESRGAN**
+([xinntao](https://github.com/xinntao/Real-ESRGAN), BSD-3-Clause). Credit: xinntao.
 
 ## Auto-rig skeleton prediction (UniRig)
 
-`unirig/encoder.onnx`, `unirig/decoder.onnx`, `unirig/embed.onnx` — ML skeleton
-prediction for unrigged meshes. These are **ONNX re-exports** of
-**[VAST-AI/UniRig](https://huggingface.co/VAST-AI/UniRig)** (SIGGRAPH 2025 — MIT
-code + MIT weights, trained on Articulation-XL2.0 / CC-BY-4.0). Converted with
-`scripts/export-unirig-onnx.py` in the QtMeshEditor repo. Downloaded into
-`<AppData>/ai_models/unirig/`. Credit for the weights: VAST-AI-Research.
+`unirig/{encoder,decoder,embed}.onnx` — autoregressive skeleton prediction for
+unrigged meshes. ONNX re-export of the skeleton stage of
+**[VAST-AI/UniRig](https://huggingface.co/VAST-AI/UniRig)** (SIGGRAPH 2025,
+MIT code + weights). Credit: VAST-AI-Research.
+
+## ML skin weights (SkinTokens / TokenRig)
+
+`skintokens/` — five ONNX graphs + manifest; QtMeshEditor's **default
+skinner**. ONNX re-export of **VAST-AI SkinTokens/TokenRig** (MIT code +
+weights, Qwen3-0.6B backbone). `decoder.onnx.data` holds the LM weights as
+external data (ORT can't parse the >1.6 GB single-file proto). Credit:
+VAST-AI-Research.
+
+## Image → 3D
+
+- `triposr/` — **TripoSR** (Tripo AI + Stability AI, MIT): triplane encoder
+  (fp32 + int8 tiers) + per-point density/colour decoder.
+- `triposg/` — **TripoSG** (VAST-AI, SIGGRAPH 2025, MIT): DINOv2 image
+  encoder, rectified-flow DiT step graph (fp32 external weights; the int8
+  tier here is deprecated — it degrades to blobs over the CFG flow loop),
+  VAE latent + field-decoder graphs. Geometry-only; colour comes from
+  TripoSR's colour field.
+- `rembg/u2net.onnx` — **U²-Net** saliency for background removal
+  (Apache-2.0, the rembg model).
 
 ## Animation in-betweening (RMIB) — trained by us
 
-`inbetween/rmib.onnx` — fills the gap between two keyframes with smooth
-intermediate motion. **Trained from scratch** by the QtMeshEditor project on the
-permissive [CMU Graphics Lab Motion Capture Database](http://mocap.cs.cmu.edu).
-Beats spherical-linear interpolation by >2× on held-out CMU motion.
-**Dedicated repo:** [fernandotonon/QtMeshEditor-rmib-inbetween](https://huggingface.co/fernandotonon/QtMeshEditor-rmib-inbetween).
-Downloaded into `<AppData>/ai_models/inbetween/`. License: CC-BY-4.0.
+`inbetween/rmib.onnx` — fills the gap between two keyframes. **Trained from
+scratch** on the permissive
+[CMU MoCap database](http://mocap.cs.cmu.edu); beats slerp by >2× on held-out
+CMU motion. License: CC-BY-4.0.
+
+## Text-to-motion — trained by us (experimental)
+
+`motion/t2m.onnx` + `motion/t2m-vocab.json` — CVAE transformer, text keyword →
+22-joint world-frame clip; `motion/motion-library.json` — the curated CMU
+template-clip library that is the shipped default. License: CC0-1.0.
 
 ## Mesh part segmentation — trained by us
 
-`segment/meshseg.onnx` — predicts head / torso / left+right arm / left+right
-leg labels per point (PointNet++-style, two kNN aggregation blocks). **Trained
-from scratch** (v2) by the QtMeshEditor project on **surface-sampled synthetic
-bodies we own** (humanoid incl. chibi, quadruped, biped-with-tail — exact
-by-construction labels, CC0) mixed with **CC0 rigged characters mined for
-exact rig-derived labels** (Quaternius packs). **94.7%** per-vertex accuracy
-on rig-truth eval (v1: 31.5%). **Dedicated repo (full model card + eval data):**
-[fernandotonon/QtMeshEditor-mesh-segmentation](https://huggingface.co/fernandotonon/QtMeshEditor-mesh-segmentation).
-Downloaded into `<AppData>/ai_models/segment/`. License: CC-BY-4.0.
+`segment/meshseg.onnx` — per-point head/torso/arm/leg labels
+(PointNet++-style). Trained on synthetic bodies we own + CC0 rigged
+characters (Quaternius). 94.7% per-vertex accuracy on rig-truth eval.
+License: CC-BY-4.0.
+
+## Image captioning
+
+`caption/SmolVLM-500M-Instruct-Q8_0.gguf` + `mmproj` — quantized
+**SmolVLM-500M-Instruct** (HuggingFaceTB, Apache-2.0) for llama.cpp-based
+captioning. Credit: Hugging Face TB.
 
 ---
 
 These models power the AI-assisted authoring features in
 **[QtMeshEditor](https://github.com/fernandotonon/QtMeshEditor)** and its
-companion **QtMesh Cloud** ([qtmesh.dev](https://qtmesh.dev)). Each downloads on
-first use and runs locally (offline). Mixed licenses per model as noted above
-(CC0 / BSD-3 / MIT-derived / CC-BY-4.0); see each section + the QtMeshEditor
+companion **QtMesh Cloud** ([qtmesh.dev](https://qtmesh.dev)). Provenance and
+licensing decisions are documented in the QtMeshEditor repo's
 `THIRD_PARTY_AI_MODELS.md`.