Skip to content

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.

5.5 KB
Raw Blame

Turn Photos into a 3D Model

Have a real object you want in 3D? Photograph it from a few angles, drop the photos into the Forge, and get back a textured 3D model. This works for products, toys, sculptures, furniture, sneakers — anything you can put on a table and walk around.

What you'll make: a downloadable GLB model reconstructed from 1 to 6 photos of a real object.

Prerequisites: none. A phone camera is plenty.

The image-to-3D page with its upload dropzone

Drop a photo here and the same pipeline returns a mesh instead of a render.


Step 1 — Take good photos

This step decides 90% of your result quality. Five rules:

  1. One object, plain background. Put the object on a table against a wall, or hold it against the sky. Busy backgrounds get reconstructed as part of the object.
  2. Even lighting, no harsh shadows. Indirect daylight is ideal. Avoid a phone flash — it creates hotspots that become weird texture patches.
  3. Fill the frame. The object should take up most of the photo, in focus.
  4. Same object, different angles. The core set is front, back, left, right; two more slots take a top shot and a three-quarter shot. Keep the object the same way up in every shot.
  5. Don't move the object between shots — walk around it instead, so the lighting stays consistent.

One photo works. Four to six work noticeably better: every angle you provide is geometry the model doesn't have to guess.

Formats: PNG, JPG, or WebP, up to 8 MB each.


Step 2 — Open the Forge in photo mode

Go to three.ws/forge and switch to the From photos tab.

You'll see a grid of six upload slots (front, back, left, right, top, and three-quarter). Drag your photos in, or click a slot to browse. After uploading you can drag the thumbnails to reorder them: put your best, clearest shot first.

No object on hand? Draw it. The Forge also takes drawings, two ways. On this page, the From a sketch tab (it appears when the sketch engine is live) takes one drawing plus a short required description of what it depicts: dark strokes on a light background work best, and the result is untextured geometry you can then Stylize or Retexture. In the Forge Studio, click Or draw it instead below the photo grid to open an in-browser drawing canvas: draw with your mouse, finger, or stylus, then Use sketch to drop the PNG straight into the next open slot, where it rides the same reconstruction pipeline as an uploaded photo. Nothing is uploaded while you draw, only when you click Use sketch; cancelling discards it with zero network activity.


Step 3 — Add guidance (optional, but worth it)

Below the photo grid there's an optional text field. Use it to tell the engine things the photos can't:

matte plastic, neutral pose, no background

Good guidance mentions the material ("brushed aluminum", "soft fabric") and anything to ignore ("no background", "ignore the stand"). Keep it short — the photos are doing the heavy lifting.


Step 4 — Pick a tier and generate

The same three quality tiers from text-to-3D apply:

  • Draft — fast preview to confirm your photos work.
  • Standard — the default.
  • High — maximum geometry + PBR textures, for the final asset.

Click Generate.

If the Forge flags your image: before spending a generation, the Forge runs a quick vision check on your photos. If it warns you — blurry, busy background, multiple objects — it tells you exactly what's wrong. You can fix the photo (best results) or click Generate anyway to override the check.

Reconstruction takes about a minute on Standard. You'll see the progress steps as it works.


Step 5 — Inspect, download, share

The finished model lands in the 3D viewer:

  • Drag to rotate, scroll to zoom, F for fullscreen turntable.
  • Check the back and the side you photographed least — that's where the engine had to guess. If a guessed area matters, take a photo of that angle and regenerate.
  • Download GLB to save it, Share for a link with a preview card.
  • Rate it 👍 Keep or 👎 Discard — your verdicts train the engine selection.

Everything you generate is kept in Your creations at the bottom of the page, tied to your browser.


Troubleshooting

Symptom Likely cause Fix
Background melted into the model Busy backdrop in the photos Reshoot against a plain wall, or note "no background" in the guidance field
Texture has bright patches Flash or hard directional light Reshoot in even, indirect light
Back of the model is wrong Only front photos provided Add back, side, and three-quarter views, up to 6
Object came out warped Photos show the object in different positions/orientations Keep the object still; you move, not it
Upload rejected File too large or wrong type PNG/JPG/WebP, max 8 MB each
"Generation limit reached" Per-visitor rate limit Wait a minute, try again

What's next