Trosfy
UserLocal generative-AI as Claude Code plugins for NVIDIA DGX Spark — FLUX.2 images & Hunyuan3D image-to-3D via ComfyUI, with gated S3/CloudFront downloads (Terraform).
Categories
Indexed Skills (4)
model
Turn an image into a 3D model (GLB mesh) locally on the DGX Spark using Hunyuan3D-2.1 via the ComfyUI container. Use whenever the user wants to create, generate, or reconstruct a 3D model, mesh, or .glb from a picture — e.g. "image to 3D", "make a 3D model of this", "generate a mesh", "photo to 3D", "turn this into a glb". Optionally decimates to a low-poly target. Runs on local weights, no cloud.
publish
Publish/export a local file or result off this headless box to the forge S3 + CloudFront store and return a shareable download URL. Use whenever the user wants to get a file out of the box — "publish this", "share this file", "upload to the CDN", "get this off the box", "give me a link for this", "export this result", "send this somewhere I can download it". Returns a signed (gated) or direct URL.
cdn
Provision the headless download CDN on AWS — a private S3 bucket fronted by CloudFront, optionally with signed/gated URLs and a custom domain — via the bundled Terraform module, then wire the forge config. Use whenever the user wants to set up, provision, or deploy the download/publishing infrastructure, e.g. "set up the CDN", "provision downloads", "deploy the S3/CloudFront", "set up headless downloads", "enable signed/gated downloads". Plans first and never applies without explicit confirmation.
hunyuan
Provision Hunyuan3D-2.1 (image-to-3D) inside the ComfyUI container so /forge:model works — clones the repo and installs the Python deps; the ~7GB shape weights auto-download on first use. Use when the user wants to set up, install, or enable image-to-3D / Hunyuan3D / the 3D model generator, or needs to re-provision after rebuilding the ComfyUI container.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.