Guide
What makes a good reference image for image to 3D?
A good reference shows one subject, whole and in focus, against a plain background in soft, even light, from a slight angle rather than flat on. The clearer its outline, the less guessing image to 3D in Hydrilla AI does.
The model can only work from what the picture contains. Every shadow, reflection, and cropped edge is information it has to interpret, so a few minutes spent on the photo usually saves several rounds of regeneration.
Subject, framing, and light
Let the object fill most of the frame, with a small margin on every side. A three-quarter angle shows two faces and a sense of depth. Avoid hard shadows and strong reflections, which can read as shape. A sharp phone photo is enough; you do not need a studio, only a plain wall and light from a window on an overcast day.
One chair, three-quarter view, plain wall, soft daylight, nothing cropped.
The same chair cut off at the legs, in a busy room, under hard window light.
The sides the camera can't see
A single image can't show the back of an object, so the model has to infer it. When the hidden side matters, pick the angle that shows the faces you care about most. For objects that look very different from behind, such as a chair with a carved back, photograph the side that carries the most detail.
Why the background matters
A busy background competes with the subject for the model's attention, and an object that matches its background loses its outline. Choose a plain surface that contrasts with the subject: a light wall behind a dark object, or a dark cloth behind a light one. Avoid colored light, which can tint the textures the model produces.
Before you upload
- One subject, with nothing touching it
- Sharp at full size, without blur or heavy compression
- The whole object in frame, with every edge visible
- A plain background that contrasts with the subject
- Even light, without blown highlights or colored casts