[X3D-Ecosystem] Does this AI program generally look correct?

Tue Jan 21 09:34:02 PST 2025

>  I found PivotConstraint by looking at the API based on what Joe told me, not anything an AI told me,  the AI was wrong.  There was no PivotConstraint in export or import so we were on new ground. 

So, in order to import/export 'standard' x3d humanoid to/from blender  ,
we need a blender transform matching an X3D Joint.
with translation 0 0 0, rotation 0 0 1 0 and
x3d Joint center equivalent to blender PivotConstraint
? 
Thanks, .
Joe

-----Original Message-----
From: Michalis Kamburelis <michalis.kambi at gmail.com>
Sent: Jan 20, 2025 10:32 PM
To: John Carlson <yottzumm at gmail.com>
Cc: Joe D Williams <joedwil at earthlink.net>, X3D Ecosystem public discussion <x3d-ecosystem at web3d.org>, Katy Schildmeyer KS APPAREL DESIGN <katy at ksappareldesign.com>, Vincent Marchetti <vmarchetti at kshell.com>, Carol McDonald <cemd2 at comcast.net>, GPU Group <gpugroup at gmail.com>
Subject: Re: Does this AI program generally look correct?

John,

My point was to suggest to you a better way, that will allow you to achieve what you want *easier*, not *harder*. 

Your comparisons indicate that you find my suggestions (read the docs, start with simple code snippets, use Blender as a regular user) harder, but I still argue that your approach in this thread ("ask AI for a larger solution and then wonder how/why the AI answer is wrong") is harder than what I suggest. 

You don't need to run the marathon, read academic documents or do other things perceived as "hard" that you mentioned.

To recap / reformulate my 3 suggestions:

- Trust Blender Python API docs (more than what AI tells you). You are reading things, you are obviously reading emails and you're reading the AI output, so I hope you will not have trouble reading the Blender Python API docs. You naturally don't need to read it "from beginning to end" like a book, just consult it as necessary. When you wonder about using X, look up what X does, follow links from X to other classes and related methods -- Blender Python API docs are heavily inter-linked for this reason.

- Try simplest code snippets in Blender. You mentioned yourself you like code snippets, this aligns with my suggestion. Try to do simple things, 1 line, few lines, see if it does what you think, then build a bigger program from these blocks. This *will* achieve a better outcome than asking AI to come up with a bigger solution.

- Do use Blender as a "regular user" and see in Blender's Python console what your (interactive) actions translate to in Python. It's really easy. You learn Python API this way. And you will then know what your users (which are Blender users in this case) actually do.

Regards,
Michalis

pon., 20 sty 2025 o 22:11 John Carlson <yottzumm at gmail.com (mailto:yottzumm at gmail.com)> napisał(a):
I’m not really sure what you’re saying Michalis.  I’ve got working gramps export code, so it would seem the task is to do the opposite using similar API on  import.  I can stare at the API for ages.  I found PivotConstraint by looking at the API based on what Joe told me, not anything an AI told me,  the AI was wrong.  There was no PivotConstraint in export or import so we were on new ground.

No, I can’t consume large amounts of documents on the web, I mostly rely on code snippets.  If I could read academic documents, I could probably have gotten a PhD or Masters.  Back when I could read better, my professors invited me to get a higher degree.

Please be considerate and realize there’s neurodiversity in what people can accomplish.  Just because I can master the bash shell doesn’t mean I can use the Blender GUI or play a 3D game to my satisfaction.  I do win world records regularly on Solitaire4us on Roku. I don’t expect people to learn vim.  I recommend notepad or vs code.  I did find this thing called vim3d which is very intriguing.  I did try a “Blender” introductory game, but it was very short.

Do I expect you to run a marathon or be a concert violinist?

What I probably should do is grab the Blender source code so i can use bash tools to learn it.  Weird why we didn’t figure that out a long time ago.  That’s how I got as far as I did with CGE.  I did look at Blender RNA a bit.

UTSL!

Ps:

I recently found “grep -L”, should be really useful!  I’ve been wanting that for quite a while!

I have found AI useful for creating a prototype Python database filtering GUI, fun stuff and no coding needed except to load the database from X3D JSON.   AI whipped out most of a naylib to Three.JS conversion as well, and the result looks better than my hand-coded stuff, probably due to API limitations.

Just because people are in 3D graphics work doesn’t mean they like to be immersed in virtual 3D.  My particular bent is mathematical visualization.  I’m not seeing my surfaces in Blender yet, I still want to my surfaces into Blender, but that may mean extending X3D or Blender.   I hired someone to get my surfaces into Blender, and they failed, despite having all my working X3D source code and shaders.

John 

On Mon, Jan 20, 2025 at 3:51 AM Michalis Kamburelis <michalis.kambi at gmail.com (mailto:michalis.kambi at gmail.com)> wrote:
Please don't rely on AI like this. Analyzing why AI said this, not that, is often a waste of time, in my experience.

AI is guessing. But you should not be guessing. You should:

- Look at documentation, of everything related (Blender Python API in general). 

- Walk step by step, testing snippets of code, from smaller (1-line) to larger, making sure your code does what you want in each iteration. 

- Do the operations in Blender, looking in Blender's Python console what your interactive operations translate to in Python. This is a great way to learn Python API.

To be clear, AI can be a big help with coding. I'm using it too. But not like this. You will just waste time trying to analyze why AI is guessing wrong ("hallucinating"), trying to find some "nuggets of truth" in what AI tells you. Instead, rely on documentation and do own research. 

Basically, use AI to help you with *your own* solution. But don't just "take a solution from AI" and don't waste time trying to analyze why AI said something wrong.

Regards,
Michalis

pt., 17 sty 2025 o 18:45 John Carlson <yottzumm at gmail.com (mailto:yottzumm at gmail.com)> napisał(a):
Looking at the screen capture some more, there’s tons of evidence that the file got loaded, then the X3D got imported, with no cleanup in-between.

Just confirming my intuition.

John

On Fri, Jan 17, 2025 at 11:13 AM John Carlson <yottzumm at gmail.com (mailto:yottzumm at gmail.com)> wrote:
It appears that import is messed up.  There are 2 armatures with gramps, when there's only one humanoid in  the file. Weird!  Looking at gramps in castle, everything looks good except for lighting.  I need to look at import code a bit more. 

I am unsure I am able to import more than one armature.  There are very strange errors happening, which I can't currently solve.

John

On Fri, Jan 17, 2025 at 9:44 AM John Carlson <yottzumm at gmail.com (mailto:yottzumm at gmail.com)> wrote:
That’s what I’ve been doing with Joe Kick for nearly a year now.  I will try to import gramps and re-export to see what’s up, now that I am more confident with export.  There could be an issue with interpolators on import, converting to NLA Tracks.

Also, some of Vince and Michalis’ .blend models can use the same treatment.

I hear you Joe, but you’re just confirming that I am doing things correctly.  We just need to compare first export with second export.  I don’t quite know what a Switch will do on import currently, should be interesting.

I guess there might be some interpolators which are not connected to TimeSensors which could be getting included in the imported animation, but importing an export should clear that up.

Hmmm.  I wish my brain worked better.

John

On Fri, Jan 17, 2025 at 3:34 AM Joe D Williams <joedwil at earthlink.net (mailto:joedwil at earthlink.net)> wrote:
Why not show what this produces.
Does not look complete but form for Joint node looks correct 
with center and skinCoordIndex and skinCoordWeights. 
Thanks,
Joe 

-----Original Message-----
From: John Carlson <yottzumm at gmail.com (mailto:yottzumm at gmail.com)>
Sent: Jan 16, 2025 10:42 PM
To: X3D Ecosystem public discussion <x3d-ecosystem at web3d.org (mailto:x3d-ecosystem at web3d.org)>, Katy Schildmeyer KS APPAREL DESIGN <katy at ksappareldesign.com (mailto:katy at ksappareldesign.com)>, Joe D Williams <joedwil at earthlink.net (mailto:joedwil at earthlink.net)>, Michalis Kamburelis <michalis.kambi at gmail.com (mailto:michalis.kambi at gmail.com)>, Vincent Marchetti <vmarchetti at kshell.com (mailto:vmarchetti at kshell.com)>, Carol McDonald <cemd2 at comcast.net (mailto:cemd2 at comcast.net)>, GPU Group <gpugroup at gmail.com (mailto:gpugroup at gmail.com)>
Subject: Does this AI program generally look correct?

If we can agree that this generally looks ok, I will try to implement this in the Blender importer:  https://claude.site/artifacts/7c4c15f2-1bde-4da5-9503-f4f25b09016a

I realize the bone.tail should probably be the child joint center.

I do not yet know what i am doing wrong, but my focus will be shifting to animation import after export is done to my satisfaction.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://web3d.org/pipermail/x3d-ecosystem_web3d.org/attachments/20250121/de61fc3e/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 234250 bytes
Desc: not available
URL: <http://web3d.org/pipermail/x3d-ecosystem_web3d.org/attachments/20250121/de61fc3e/attachment-0001.png>