How to choose a reference image for image to 3D

Image-to-3D has to answer a hard question: what does the rest of this object look like? A single picture shows one side. Everything else, the depth, the back, the underside, is inferred. The better the picture, the less the model has to guess. This guide covers what to look for before you upload.

1. One object, fully in frame

Crop so the whole object is visible with a small margin on every side. If part of it is cut off by the frame, the generator will either cut it off in 3D or invent something. Remove other objects from the shot; a cup next to a teapot will often get fused onto it.

2. Soft, even light

Hard light creates dark shadows and bright highlights. The generator cannot tell a shadow from dark paint, so shadows end up baked into the texture and look wrong once the model is lit from another direction. Shoot near a window on an overcast day, or use a light tent. For product photos, the flat, even lighting of catalogue shots is ideal.

3. A three-quarter angle

A straight-on front view hides depth: a box and a flat panel look the same. Turn the object about 30 to 45 degrees so you can see two sides and a bit of the top. That single change gives the generator far more information about thickness.

Exception: perfectly symmetric objects, like the bottle in the image to 3D demo, work fine from the front.

4. Avoid wide-angle distortion

Phone cameras held close to an object exaggerate perspective: the near parts look huge and the far parts shrink. That distortion becomes part of the 3D shape. Step back and zoom in a little, or use the phone’s 2x or 3x lens, to get a flatter, more accurate view.

5. Enough resolution, no heavy compression

At least 1024 pixels on the short side is a good minimum. Screenshots of thumbnails and images saved many times as JPEG have blocky artefacts that turn into lumpy surfaces. Use the original file where you can.

6. Plain background

Background removal is automatic, but it works best when the object stands out clearly. A plain wall or sheet in a contrasting colour is ideal. Avoid backgrounds that are the same colour as the object, and watch for transparent or reflective parts that the background shows through.

When to use more than one view

Add side and back views when the parts you cannot see are different from the parts you can. Typical cases:

• Backpacks, bags and clothing, which have straps, zips or prints on the back
• Chairs and furniture, where the back and legs are hidden from the front
• Characters with tails, wings or detailed hair
• Vehicles, whose sides and rear are completely different from the front

Shoot all views at the same distance and lighting, and keep the object in the same position. Rotating the object rather than walking around it is fine, as long as nothing else changes.

Drawings and concept art

Clean, finished concept art converts better than rough sketches. A few tips specific to drawings:

• Draw a single object in a neutral pose on a blank background.
• Use flat or soft shading; dramatic rendered lighting causes the same baking problem as photos.
• Close all outlines. Gaps in line art can become holes in the mesh.
• Add a short text hint with the materials, because a drawing does not always make them obvious.

Check your rights before uploading

Only use images you made, own, or are licensed to use. A 3D model made from someone else’s product photo or character art still copies their design. The licensing page explains what you own when you generate a model.

Quick checklist

• Whole object visible, nothing else in frame
• Soft light, no hard shadows
• Three-quarter angle unless the object is symmetric
• Shot from a distance with a slight zoom
• 1024 px or more, original file
• Plain, contrasting background
• Extra views if the back is different