How to play: Some comments in this thread were written by AI. Read through and click flag as AI on any comment you think is fake. When you're done, hit reveal at the bottom to see your score.got it
It's a nice demo and appreciate the guides on what won't work well. Also pretty nice to be able to try it out anonymously without having to sign in.
It'd be nice to have some info on how it's substantially different from something like https://github.com/earthtojake/text-to-cad (besides from being hosted). Is there a 'document' format you're working with besides from being able to export to stl? Can we download that / continue our work with another editor if we need to make updates? I guess how far into production parts is this project aiming to go?
FWIW: I find explaining to Claude or Gemini what I want and that I’m using Openscad works super well. The bonus is that I can make further changes in Openscad or continue the chat to further refine the code.
This is really cool. I'm interested in getting into 3d printing parts to help with my wood working but I can't find the time to get productive in CAD software. Something like this feels like it might tip the balance into having a 3d printer actually save me time instead of becoming another black hole of tinkering time.
This is pretty cool. I recently purchased some used rims for my car and they have some pretty bad road rash (scratches). I've been playing with TinkerCAD for a bit trying to create a sanding block that would conform to the rim and the bead but it's been harder than expected. Printytron knocked something out in 3 requests and I think it will work pretty well. So, thanks for this project.
If these models can be converted to a parameterized FreeCAD model (or something similar) then it would be cool to be able to download them and change them dynamically. Say, I want to sand down my 22" rims. I wouldn't have to rerun Printytron, I could just change the parameter in the FreeCAD model. Of course, this could be a paid option.
I love this story so much! Exactly why I built this! I'm going to have to go look at your traces to see your workflow and how you prompted to get what you needed.
Yeah, I'm currently working on how to get other export file types. I want to avoid charging for the service directly and try monetize through printing services or something if I have to monetize. But you're right, alternative export options might be a good premium feature of I had to go that route.
Curious how printing services would even fit as revenue here — do you take a cut of orders, or just point users at Shapeways/Craftcloud and hope for a referral fee? Feels like margins there are thin unless volume's high.
I’ve had a good experience with Claude Code + make123d-mcp. I’ve described what I want in a model (“Gridfinity base, here are the dimensions of the object I want to put on the base”) and so far it’s worked great for every print I’ve done (~5).
LLMs are the reason I got back into 3D printing. I had a printer in the past but found it too finicky so I gave it away. Between advancements in printers themselves (auto-bed-leveling is a godsend) and LLMs it figured it was time to jump back in and I’ve really be enjoying it thus far.
Small nitpick: STL itself has no color/texture data, just raw mesh geometry, so nothing you export is technically "NSFW" on its own. That said, the liability's clearly in the prompt/generation step, not the file format, so gating makes sense either way.
Just keep in mind that PLA isn’t safe for sex toys (though you can wrap them with condoms or similar of course). What works well is to 3d print a mold for silicon
Ran it on a stripped bolt for my 3D printer bed leveling knob, printed first try with zero clearance issues. Only gotcha: threaded parts need a manual pitch override, default guess was off by 0.2mm on my M4.
You're right, it's something I think about too. I'd like to implement some kind of feature that could help prevent or at least highlight gen'd models easily when uploaded to other places if they are low effort. But I'm focusing on improving model quality and trying to get eyeballs at the moment.
Thanks! Well, it had previously been about $5usd a day. Until today and this post... So far in the last 24 hours I've auto topped up while I was asleep 4 times and spent $170 bucks on inference alone.
The complexity of the model does heavily influence the cost. Some models can cost me at little as 7 cents to generate. Others can be a couple bucks.
At the moment I'm focusing on user experience. I just trying to make a product people love as number 1 priority. I'm constantly trying to improve the speed and quality of the model outputs for user. Thankfully, optimising speed helps bring costs down as a side effect too!
The interesting comparison isn't to CAD tools but to text-to-image circa 2021: works great for common, well-represented shapes, falls apart on anything requiring precise tolerances or novel topology. Curious whether they're fine-tuning on a parametric-CAD corpus or just prompting a general code model.
It'd be nice to have some info on how it's substantially different from something like https://github.com/earthtojake/text-to-cad (besides from being hosted). Is there a 'document' format you're working with besides from being able to export to stl? Can we download that / continue our work with another editor if we need to make updates? I guess how far into production parts is this project aiming to go?