Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
NVIDIA Inception membership
three.ws is a member of the NVIDIA Inception program, NVIDIA's global program for startups building on accelerated computing. Membership was accepted in July 2026.
This page explains what that means in practice: what already runs on NVIDIA, what membership adds, and where it shows up in the product.
Why NVIDIA is a natural fit
Every 3D generation lane on three.ws already runs on NVIDIA silicon:
- Text-to-3D and image-to-3D (the /forge free lane and the paid lanes) generate on NVIDIA GPUs on Cloud Run.
- Photo-to-avatar reconstruction, auto-rigging, and motion capture run on the same GPU fleet.
- LLM chat, vision, embeddings, speech, and facial animation run on NVIDIA-hosted models. The full model-by-model map is in NVIDIA models on three.ws.
GPU capacity is the single hardest constraint on how fast and how detailed generation can be, which is exactly the constraint Inception helps relax.
What membership adds
- GPU credits and hardware access — more capacity for avatar generation and open-model inference, on top of the free-first lanes.
- Technical guidance — migrating the digital-human stack toward NVIDIA ACE (Audio2Face + Riva) and NIM / TensorRT-LLM.
- Ecosystem access — Omniverse/OpenUSD interop, co-marketing, the NVIDIA VC Alliance, and the wider Inception network.
Membership is a startup program, not a partnership or an investment: NVIDIA reviews applicants and admits companies building seriously on accelerated computing.
Catalog and event consideration
NVIDIA clarified the operating path on 2026-09-14. Accelerated Application Catalog consideration comes from a periodically reviewed Inception product record, not an automatic publication button. A qualifying record must describe NVIDIA technology used at runtime, be set to Shipping, distinguish technology currently used from technology only being considered, include supporting Product Description and Technical Details, and carry a product logo and brand color.
The GTC Startup Pavilion is separate and invite-only. NVIDIA identifies potential candidates from current member profiles, clear demos of NVIDIA acceleration, and ongoing work with NVIDIA contacts. A public GTC call for submissions, when open, is a separate speaking/poster route and should not be described as a Pavilion application.
Where you'll see it
- The NVIDIA Inception member badge in the site footer, linking to NVIDIA's startup program.
- The membership overview deck: the platform, the numbers, the team, and what we're doing with the program.
- The changelog, where membership was announced on 2026-07-21.
Related
- NVIDIA visibility map: every NVIDIA surface where three.ws can earn recognition, what each one requires, and which membership benefits are still unclaimed
- NVIDIA models on three.ws — the free inference layer, model by model
- The generator was never the hard part: our Nemotron Nano write-up, published on the NVIDIA Developer Forums
- Image-to-3D on NVIDIA L4 and Blackwell: the engineering post on the GPU fleet itself: memory ceilings,
sm_120kernels, and regional quota - How the forge works — the 3D generation pipeline the GPUs power