Tucson Sculpture Festival 2012

▼

Tuesday, 8 September 2026

AI Free Form

I used the new edit approach to "combine images" and further edit them in Midjourney 8.2 to generate the surreal fantasies below:



































GAUSSIAN SPLATS
***
2D to 3D  -- Holograms -- Augmented Reality


2D to 3D

I converted some of the above Midjourney images into 3D Gaussian Splats with TripoSplat in ComfyUI, and combined them in SuperSplat to make the 3D composition below:

(13.62 MB slow loading)
Download the individual Gaussian SPZ files:
  260910SPZ.ZIP (8.91 MB)


GAUSSIAN to pseudo-HOLOGRAM

I would love to make a hologram with a 3D Gaussian Splat.  So I did the next best thing, and put a 3D Gaussian Splat on the Looking Glass Go Hologram display: 

I started with this 2D AI image

I first converted the 2D image into a 3D Gaussian Splat in ComfyUI using TripoSplat -- in PLY format (not SPZ) -- as displayed in the SuperSplat scene below: 

(6.79 MB slow loading)

I then uploaded that PLY Gaussian Splat into G Splat Explorer from Looking Glass -- to convert it into a PNG to then upload to BLOCKS: 


Finally I then transferred that "hologram PNG" from BLOCKS to the Looking Glass Go hologram display:

I uploaded the PLY Gaussian Splat
and viewed/transferred it to

The Gaussian Splat image

  • Ultimately it would have been easier just to make the Looking Glass hologram from the original 2D image, rather than first converting it into a Gaussian Splat.

LINKS:

GAUSSIAN to AUGMENTED REALITY
(workaround)

I generated a PLY Gaussian Splat in TripoSplat in ComfyUI, and then converted that into a GLB file with Spalt2Mesh.  Finally I used <model-viewer> to make an AUGMENTED REALITY scene:


    • The TripoSplat in ComfyUI workflow has an experimental convert to GBL option, but that leaves a lot of holes in the 3D object. Spalt2Mesh seems to work better.

ANIMATION

For fun, I like pushing AI further and animating the Augmented Reality screen shots in Midjourney:

Animated AR Screen Shot


***

GPT 6 ASTRA came out publicly on September 4, 2026:

  • ai-toolkit is claimed to be the best program for training LORAS.

NOTE:  We revived the AI computer at Quelab -- as some stealth crypto mining programs were hogging up all the resources, and preventing CompyUI from generating Gaussian Splats for me.  Adam took off "Nvidia mining gpu(?)" first.  And on September 9, 2026 Adric took off two instances/folders of XMRig (one installed on August 13th and the other installed on August 22nd).  Apparently they were installed in the CompyUI folder as they needed a Python script to work.  Also apparently ComfyUI can be hacked to mine crytocurrency.


PROJECTION

Now if only I can project my AI BIG:


Morgan Barnard 
speaking at the AMBIENT INTELLIGENCE exhibition

UPDATE (September 16, 2026): We projected some of my AI ART images outdoors at Quelab with the powerful projector:
 
Projected on the back wall
of the Quelab parking lot

The projector was not high enough
to project above the fence

We projected from a low table

The projection looked great
inside the half dome EV charging station

In short, the AI ART images in this blog post look great when projected large.  The same can't be said of my figure drawings:





No comments:

Post a Comment

Note: only a member of this blog may post a comment.