I used Google AI Studio to generate a 3D objects from a 2D images -- using the "Gemini 2.5 Pro Experimental 03-25" model (for free) -- by generating code for OpenSCAD (though the results are not always pretty):
- (YouTube): Gemini 2.5 is so cracked
2D into 3D using the model
"Gemini 2.5 Pro Experimental 03-25"
Animation of all 5 figures
generated by
"Gemini 2.5 Pro Experimental 03-25"
BOCCIONI
I downloaded the famous sculpture by Umberto Boccioni -- Unique Forms of Continuity in Space -- distorted it, and then used AI to regenerate that 2D distortion into a 3D figure:
DOWNLOAD: Boccioni3D.GLB (440.36 KB)
After distorting Umberto Boccioni's sculpture with SculptGL and Midjourney -- I uploaded the 2D result to Google AI Studio and generated the OpenSCAD code with the Gemini 2.5 Pro Experimental 03-25 model:
Then I uploaded that code into OpenSCAD Playground online and generated a 3D GLB object:
GLB file generated online
The result was not so pretty, but it was amazing that I could generate a 3D printable object from a 2D image just by using an LLM (Large Language Model).
REDMAN
Rinse and Repeat. I used the same process to generate a new 3D object from a figure that I originally drew in Paint 3D in 2021:
DOWNLOAD: Redman.GLB (628.21 KB)
GLB file generated online
FROG MODEL
Again, I turned a 2D screenshot of a 3D figure that I drew in 2021 into a similar 3D figure using Google AI Studio, and then even put that 3D result into AUGMENTED REALITY:
DOWNLOAD: Frog3D.GLB (146.44 KB)
GLB file generated online
SITTING
I also took another 3D figure I drew in Blender in 2021, converted the 2D image into 3D again using Google AI Studio, and 3D printed it at Quelab:
Ally 3D printed the above figure for
DOWNLOAD: Sitting.GLB (49.03 KB)
GLB file generated online
3D PRINT
The beginning of the 3D print
From AI into the Physical World!
The next step would be to VACUUM FORM a mold from the 3D print, like we did in 2020:
JAVELINAS
DOWNLOAD: javelinaorigami.GLB (2.5 KB)
Google AI Studio labeled my
javelina as a "boar"
There were errors in the code,
but OpenSCAD Playground generated a 3D javelina anyhow
FAIL
We had done this before in late December 2024 -- Text-to-3D -- when Tim asked ChatGPT to generate OpenSCAD code to create a 3D javelina at Quelab:
CHATTING up BLENDER
Imagine just chatting with AI to generate a whole 3D scene in Blender, from a 2D image. You can do that with MCP (example on X).
There is a TRIOPO plugin for Blender that will allow you to upload a 2D scene and make a 3D scene of it in Blender -- but it costs money:
- TRIPO -- cost money
MCP
- (YouTube): AI Just Took Over Design, Code… and Blender! (GPT-4o Image Gen, Blender MCP, Gemini 2.5)
- (YouTube): I gave Claude root access to my server... Model Context Protocol explained
- (YouTube): This Chrome Extension Makes AI Agents Super Easy!
- There is an AI radio station -- HIT RADIO AI -- that plays all music generated by AI
- On April 3, 2025, Midjourney just released their "V7 Alpha model," but makes you 'personalize' to use it.
No comments:
Post a Comment
Note: only a member of this blog may post a comment.