印刻万物 TOP3DGS印刻万物TOP3DGS
Things in Hand 场景封面

scene 01 · Objects

Things in Hand

掌中万物

Bring any object into the palm of your hand

Open interactive ↓

Scroll to the interactive model

Live Interactive

Drag to explore this scene

A real 3DGS model for this scene — drag to orbit, scroll to zoom, double-click to reset. Hosted by our studio rather than relying on third-party public servers.

Drag / scroll / double-click

Loading…

Model credit: Inktoys · Object capture demo

Find your path

Answer 3 questions and we'll recommend the best workflow for you.

Q01

What gear do you have?

Q02

What quality do you need?

Q03

How much time can you invest?

Or browse all three paths below ↓

Three end-to-end paths

Pick your path

From the easiest to the most professional. Each path is tagged with time, gear, and best fit.

Cloud Workflow

30-minute first run

Just your phone and an app — produce your first Gaussian Splat without installing anything.

Difficulty
Beginner
Time
30 min – 2 hr
Cost
Free quota / a few dollars per scan
Best for
First-timers, fast e-commerce listings, creators who avoid the CLI

Tools you'll use

  • Polycam
  • Luma AI
  • Kiri Engine

Step-by-step

  1. 01

    Set up the environment

    Place the object on a matte, light-colored surface with diffused lighting (cloudy window or two softboxes). Avoid reflections.

    Open the chapter · Subject Taxonomy & Capture Strategy →
  2. 02

    Orbit-shoot 80–150 photos

    Walk 2–3 loops around the subject at varied heights (low / eye / high). Adjacent shots should overlap >70%.

    Open the chapter · Site Recon & Shot Planning →
  3. 03

    Upload to the cloud

    Create a new Splat project in Polycam or Luma AI, drop in your photos or video, and wait 5–30 min for training.

  4. 04

    Trim and tune online

    Open SuperSplat in the browser to remove ground/background floaters and tune color or density.

    Open the chapter · Post-Processing, Cleanup & Compression →
  5. 05

    Export and embed

    Export .splat or .ply, share via the platform link, download locally, or embed on your own site.

    Open the chapter · Distribution & Display →
Path 01 / 3Cloud Workflow
Local GUI Tool

Reliable, controllable quality

Camera RAW + local GUI trainer — no code, noticeably higher ceiling than cloud.

Difficulty
Intermediate
Time
Half a day – 1 day
Cost
Postshot trial free / ~$120 one-time + a 6GB+ NVIDIA GPU
Best for
Camera owners, quality-obsessed solo creators, small teams shipping to clients

Tools you'll use

  • Postshot
  • Jawset
  • RealityScan

Step-by-step

  1. 01

    Shoot 200–400 RAW frames

    Lock aperture / ISO / WB. Tripod or slow walk. Shoot RAW so grading has headroom.

    Open the chapter · Camera Settings & On-Site Operation →
  2. 02

    Batch grade in Lightroom and export

    Grade one reference shot, sync to the rest, export JPEG (~3000px long edge) for SfM.

    Open the chapter · Color Grading & Consistency →
  3. 03

    Solve poses in RealityScan

    Drop JPEGs into RealityScan, run Align Images. Aim for ≥99% alignment before training.

    Open the chapter · SfM Pose Estimation →
  4. 04

    Train 30 min – 3 hr in Postshot

    New Project → import photos + poses → pick the 3DGS preset → adjust batch size to your VRAM.

    Open the chapter · Training →
  5. 05

    Polish and export in SuperSplat

    Crop bounding box, kill floaters, export .splat (smallest) or .ply (most compatible).

    Open the chapter · Post-Processing, Cleanup & Compression →
Path 02 / 3Local GUI Tool
Professional Pipeline

Cinematic, deliverable-grade

Studio lights + turntable + COLMAP/GLOMAP + train from GitHub source. Repeatable, tunable.

Difficulty
Professional
Time
1 day – 1 week
Cost
Studio lights + turntable from $500+ / RTX 4090 or rented cloud GPU
Best for
Client delivery, museum digitization, automotive / luxury visual teams

Tools you'll use

  • Studio lights + turntable
  • COLMAP / GLOMAP
  • inria-3dgs / gsplat
  • Blender / Unreal Engine

Step-by-step

  1. 01

    Studio lighting + calibrated turntable

    Three-point soft lighting, kill specular highlights. Capture every 5° at three heights — ~216 frames total.

    Open the chapter · Subject Taxonomy & Capture Strategy →
  2. 02

    COLMAP CLI reconstruction

    feature_extractor + exhaustive_matcher + mapper. Tune params to minimize mismatches.

    Open the chapter · SfM Pose Estimation →
  3. 03

    Train 300k – 500k iters from source

    Clone inria-3dgs or gsplat, tune density threshold, SH degree, and learning rate per scene.

    Open the chapter · Training →
  4. 04

    Composite in Blender / UE

    Import .ply, light the scene, add depth passes, composite with CG assets for final shots.

    Open the chapter · Post-Processing, Cleanup & Compression →
  5. 05

    Multi-platform delivery / custom viewer

    Export multi-resolution versions; build a spark / gsplat.js viewer per spec, or deliver USDZ.

    Open the chapter · Distribution & Display →
Path 03 / 3Professional Pipeline

TOP3DGS · Commercial service

Need a hand?

Let TOP3DGS shoot it for you

Full-service delivery from studio lighting to final asset, per-piece pricing, weekly turnaround. Built for e-commerce hero shots, museum digitization, and product launches.

Get in touch →The bridge between technology and your market