HN Showcase
Back to browse
Image-3D: photo to 3D splat that runs in the browser

Image-3D: photo to 3D splat that runs in the browser

by mnorris·Jun 8, 2026·2 points·1 comment
Visit ProjectView on HN

AI Analysis

●●●BangerWizardryZero to OneSlick

Browser-only 3D splat generation beats cloud APIs with WebGPU and WASM fallback.

Strengths
  • •Depth Anything V2 via ONNX Runtime Web runs inference entirely client-side.
  • •WebGPU with WASM fallback ensures cross-device compatibility.
  • •Two-line web component embed with shadow DOM and lifecycle events.
Weaknesses
  • •~1 GB RAM during inference limits use on lower-end devices.
  • •30 MB model download on first visit impacts initial page load.
Category
AI/ML
Target Audience

Web developers, frontend engineers building interactive experiences

Similar To

Luma AI · Kaedim · CSM.ai

Post Description

I'm Michael. Same demo I posted a few weeks ago, no server this time: photo to 3D Gaussian splat, all in the browser.

Depth Anything V2 Small runs via ONNX Runtime Web / @huggingface/transformers. It tries to use WebGPU (which does inference in about 1 second on an M4 Pro processor and 2 seconds on an iPhone 17 Pro) and falls back to WASM (about 5 seconds inference on the same devices).

It's a web component, two lines to embed.

Demo: https://mukba.ng/image-3d/demos/local/ Docs: https://mukba.ng/image-3d/ Source: https://github.com/imichaelnorris/image-3d

Similar Projects

Open Source●Mid

SVO Voxelization for Gaussian Splat Collisions

Using an SVO to voxelize Gaussian splats is a sensible way to prune overlap checks — hierarchical voxels fit the problem and should cut costly pairwise collisions. Can't judge the execution: the Reddit thread is blocked with no visible code, benchmarks, or demos, so this currently reads like an intriguing sketch rather than a drop-in tool.

Niche GemWizardry
slimbuck
503mo ago