I thought I would make a thread focused on AI generated height maps for cnc or 3d prints that I share. I am still learning and refining my workflow so these are not perfect but should be usable.
I am currently using a 4 step process.
Generate the AI image
Generate AI height map from image
Create geometry from height map and refine and clean the model
Manually create another height map from the geometry to maximize height of pattern (I use a logarithmic function to focus on near geometry detail and reduce distant geometry detail)
The height map generated by the AI will more than likely not be perfect or have the depth proportions that are desired.
In step 3 I bring the height map into a 3d modeling program (Blender3d) and 3d model/digitally sculpt the areas that I want changed. That could mean raising and lowering different areas of the model. I can also refine/add/delete areas for either more or less detail.
Many times I get a resulting 3d model with exaggerated proportions that is unusable for a straight STL port.
So step 4.
I make another height map using a logarithmic function to compress the model information so far depth information is reduced and and near depth information is kept.
Very simple node function to accomplish this. You do have to play with the numbers and every model will be different.
AI is helping but your talent with tools is still necessary to get usable results. That’s not surprising; AI is evolving. The panacea would be an app using an OpenAI API where uploaded images produce downloadable (and usable) 3D files. We’ll get there I expect.
I found this more challenging than I thought it would be. I spent about 15 minutes trying to clean up the model but it probably needs at least another hour and my drawing pad to control brush strength.
Great work!
I owned two English bulldogs in the past. It’s funny how reality on the left is not how everyone thinks of them. Your wonderful depiction on the right is how everyone pictures them.
