Optimising AI Generated Models for 3D Printing

Description

AI generated models tend to be uniformly dense, that is, the level of structural details tends to be the same regardless of the model. Where details is needed, for 3D printing for example, this can mean printable objects look ‘soft’, or ill-defined, despite being relatively high-resolution. In other words, AI model generation tools don’t understand structural optimisation. A way to work around this limitation is to process two data sets (images) and then combine them in Blender with the selective use of Boolean.

Duration: total c. 15 mins (00:15:00).
Info: 1080p.
Suitability: Beginner+.
Source: n/a.
Product ID: n/a.

Design note: additional considerations using AI generated 3D models for 3D printing;

Data Sets/Source Images

As 3D models generated by AI are uniformly detailed or structured, this can be taken advantage of by ensuring source data, source images, are all close to being the same dimensions, i.e. 1920 x 1080, and that the subject matter occupies a similar focus, i.e. the element to be converted is the main ‘subject’ of the image, to ensure generators are processing meshes equally. This might mean cropping, colour-masking, or otherwise removing ‘distractions’ so the AI has the cleanest image possible.

Boolean

The shape used to Boolean the unwanted elements of resulting generated meshes can be as simple or complex as is necessary in order to ‘cut’ or remove unwanted parts; a sphere suffices to remove a head, but it might need editing to remove the head and the neck without touching the t-shirt a character might be wearing. Where this is needed, as the Boolean is a simple mesh object, more complex shapes can be edited or combined together to form a single control object.

While mesh primitives are typically used as the Boolean operator, they can be more complex depending on what needs to be removed.

Posture/Pose

For characters in particular, extra care needs to be given to posture, pose or poise, to mitigate the amount of clean-up needed when matching parts together cut from figures posed differently; if both are standing upright but one has their head turned to the side, this difference affects how the two objects are then joined at the neck – the twist of the head, twists the neck, which misaligns it, which can complicate the ‘merging’ process.

For AI generated characters, pay close attention to ‘pose’ as differences can make matching parts more time consuming than would otherwise be.

Clothing

Similar to posture/pose, clothing has to be taken in to account to avoid inclusion in the Boolean section being removed, and the extra this necessitates to clean up. Wherever possible ensure models are similarly dressed.

Shown in Material Preview for clarity, the Boolean has captured part of the characters clothing, which has to be removed so parts can be joined properly.

Timestamps

Times are approximate;
– 00:00 : Overview
– 02:00 : Boolean
– 07:00 : Boolean Operation
– 10:00 : Join & Tidy Up