3DLoom3DLoom

How to prepare a photo for an AI 3D model

Use one sharp photo of a single object with the whole thing in frame, even light and a plain background. A slight angle shows more of the form than a dead-on shot. Avoid hands, other objects, hard shadows and motion blur, because the AI treats whatever it sees as part of the shape.

  • JPG, PNG or WebP
  • 15 MB
  • One object, fully in frame, plain background

New accounts get 20 free credits. Failed generations are refunded.

How it works

  1. Step 1

    Isolate one object

    Move everything else out of the shot, including your hand. If you have to hold it, put it on a table instead.

  2. Step 2

    Fill the frame without cropping

    Leave a small margin all around. A cut-off edge gets invented or flattened.

  3. Step 3

    Use soft, even light

    Shoot near a window on a cloudy day or under diffused light. Hard shadows can turn into shapes and into the texture.

  4. Step 4

    Choose a plain background

    A plain wall or sheet is enough. A PNG with a transparent background is the cleanest input of all.

  5. Step 5

    Shoot from a slight angle

    A three-quarter view shows the front and one side, which gives the AI more of the form to work from.

At a glance

File typesJPG, PNG or WebP
Largest upload15 MB
Best inputOne object, fully in frame, plain background
Hardest partsStraps, chains, hair, fingers
Multi-viewExactly four images: front, left, back, right

What goes wrong, and why

Shiny and glass surfaces reflect their surroundings, and those reflections can end up baked into the texture or misread as shape. Dull the surface if you can, or expect to clean it up.

Thin parts are hard because they cover few pixels. Photograph straps, chains and fingers clearly against the background, and expect them to come out simplified.

Busy patterns can look like surface detail. A strong printed pattern may turn into bumps in the mesh, so check the shape-only view after generating.

Preparing four views

Multi-view works when all four images show the same object in the same pose. Keep the object still and move the camera, or turn it exactly a quarter turn each time.

If you are drawing or generating the views, check where every held or sticking-out part sits in each image. In our test, a side view that put weapons in front of a character while the front view put them at the sides produced an extra weapon. Details are in our accuracy test.

A side view should show one object edge-on where the front view shows it flat. If a blade looks wide from both the front and the side, the images disagree.

Test cheaply

A failed generation is refunded automatically, but a finished model you don't like uses its credits. If an object matters, take several photos, look at them at full size, and upload the sharpest one first.

Questions

What resolution does the photo need?+

A sharp phone photo is plenty. Sharpness and clean framing matter more than megapixels, and the upload limit is 15 MB.

Should I remove the background myself?+

A plain background is usually enough. If you already have a PNG with a transparent background, use it.

Can I use a drawing or an AI image?+

Yes. Clean art with a clear silhouette and one subject works well, and it should follow the same framing rules.

Front view or three-quarter view?+

A slight angle usually gives the AI more to work with. Use a straight front view when you are also supplying side and back views.

Related tools