79 total · 79 loaded
I'm going to try https://sawdust.diy and https://fluidcad.io/ for the short term.
For the long term, I have noted FreeCAD + Blender + $ai_model to be a good general purpose workflow that yields more control, and will probably get me lost into the details.
https://janne-alatalo.fi/en/articles/silhouette-modeling/ and its accompanying SKILL - https://github.com/janne-alatalo/image-to-cad-skill
FreeCAD MCP: https://github.com/neka-nat/freecad-mcp
Woodworking specific FreeCAD resources : https://github.com/dprojects/Woodworking
Two YouTube Channels related to FreeCAD + woodworking :
1. https://www.youtube.com/@dprojects.woodworking
2. https://www.youtube.com/@ga3d_._tech528
There goes the weekend! Cheers HN!
It's a woodworking domain engine and parametric furniture builder focused on projects like bookshelves, cabinetry, and shop furniture. It works out the parts, joints, cut list, etc.
There is an MCP component in beta (https://shopspec.io/agents) that allows an agent to explore designs while the engine still does the math. To take a design to CAD, a project can be exported as DXF or as a Ruby script that rebuilds it in SketchUp Pro.
It's narrower than Shapr3D, but it might fit the projects you describe.
As an engineer obsessed with the finer details and getting his fix from weekend woodworking by making things just the way I want for fun, and tinkering with newer ways of doing things, some finer details that probably won't make a lot of rational sense, but I really badly want them because it is cool.. I want a more general purpose workflow where I can make anything I feel like... {}toys for kids, bike stand for the shed, A tool wall, Bed for the kids..}. Your approach to doing what you happened to need - bookshelves, workbench etc in a customisable way is very cool. But, I've already made those over the years and need something else.. :-)
Your work gave me a lot of ideas and levelled up what output I can make once I figure out a more general purpose workflow.
If you're CNC cutting parts that need to fit together, you need to measure both your plywood and your cutting tool with digital calipers to the thou. Freud's 1/4" upcut spirals (available at Home Depot) are pretty commonly .245" rather than .250". That's enough that things don't fit.
Yes, I've been burned by this.
Also, if you're just making through cuts on your CNC, reference the bed for the Z-axis rather than the material surface, and a few thou either way doesn't matter.
You could just say you don't like the excessive use of caps.
If AI could catalyse that first part and help visualise the idea in many perspectives that are easy to tweak and improve, I think it will be a solid base to work upon and change organically as we build it in the shop.
I have a lot of respect for the intuitive-improvisational mode, and some people can do it brilliantly while playing the violin, but I'm not sure it always works for making things.
His woodcarvings were _amazing_, in particular for his ability to work with the grain and even to chase a single ring so as to make a perfectly shaped, but seemingly impossibly thin and light wing of a bird.
I'm also the developer of FluidCAD. It has an MCP you can wire to any LLM. It runs locally on your machine.
It is still missing features for splitting the model into parts, but that is coming in a future release.
MCP config (node.js required): https://fluidcad.io/docs/cli/#fluidcad-mcp
I personally hate that pattern, and my default reaction is to immediately close the tab.
Perhaps you could change the landing page to avoid that negative first impression?
I will say the differential wheel module was on the edge of what it was able to do. I had to bump it up to Fable and if I didn't already have a clear model in my head of what I wanted I don't think it could have pulled it off.
Recently, Sonnet 5.5 got quite good, too, but not at the level of Opus/Fable for complex geometry
Disclaimer: I use it for very precise 3D printing, not woodwork.
I had it design a model sailboat hull, and it fucking nailed it. It even designed it for printability. I’m genuinely really impressed.
I learned OpensCAD manually but it was so painful because you needed to take into account the most minimal and ridiculous thing using code, so I used Freecad in the end that is way more intuitive and fast with graphical tools.
With code like OpenScad it was just too exhausting just to have to think rationally every single detail. And you spent hours doing the most minimal project.
Then LLMs appeared...
Now I use OpenSCAD a lot because with LLMs you have the absolute control that code gives you without the pain of using your logical/rational mind all the time and taking 1/20 of the time for that.
I create my own furniture as hobby. Mostly I use standard IKEA metallic pieces, standard metric screws.
I generate paper blueprints automatically of how the wood panels must be cut and they cut it for me. My Blueprints are pdfs that I print very similar to IKEA blueprints so the one that cuts the wood can not make a mistake. I choose and provide the panels myself.
It is ridiculous how easy it is now with LLMs to do all of that. I can do a new design in a couple hours, including the pdfs. Most of the code is already created and I just have to reuse it.
I use Claude Code. From time to time I voluntarily teach children to do just that. Some of them are really good after a couple hours of instruction and some mentoring.
For you specifically; if all you want to do is manually have a session and describe the thing you want, and see it take form — that's OpenAI's expertise. Pay 20$, get the desktop app, tell it what you want with as much precision as you can. Don't tell it how (don't force CAD if you don't specifically need it). Give it pictures or a photo of a drawing, any data like wood type, maybe even the machines you have at hand... Just dump it all in there. You'll be amazed.
Anyway, I'll plug India, the boat I've reconstructed and had the absolute privilege of helping design the rig for. So that's my weird hobby that AI helps with... rigging Pirates of the Caribbean-style boats.
(Yes, we actually sail them in our community. There was a race a couple of weeks ago. India's yet to be rigged, so she just tagged along as a fender against the cliffs.)
If you have some physical reference, for example a coffee table, you can create an accurate 3D model of that quickly. The agent creates model that can be converted to STEP format and opened with any CAD software.
This might help you? You get a good model that you can start editing towards your own design.
Here is agent skill that works with OpenAI models and Codex https://github.com/janne-alatalo/image-to-cad-skill
Basically you take few pictures. Three is usually enough of furnitures. The agent uses the pictures to find the exact camera poses and creates a 3D model that minimizes the difference between the object's silhouette in images and in the 3D-model when projected with the virtual cameras to the 2D planes. The agents create the model with python scripts. First using numpy for faster optimization and finally CadQuery for easy exports to CAD formats.
Does anyone know what the SOTA is for cad models from image or multiple images?
I was hoping to try doing some reproductions of stuff i saw in museums/palaces in Europe.
A simple test is that if the object you are modeling could be described by a brief python program (automating blender/CAD APIs), then the LLM will probably one shot it perfectly. If what you are modeling is a human face or bird, then the LLM is at best a clumsy apprentice. High Kolmogorov complexity is the common thread behind all things that LLMs will struggle with.
It can sometimes be easier to lean on a diffusion model and then trace over the results the old fashioned way. Diffusion completely sidesteps the geometry which helps with design speed.
Everything is stored locally (with cloud feature coming soon as well as desktop versions.)
I've been working on this sort of thing at a far lower-level (directly modeling toolpaths and tool shapes so as to get 3D models of parts).
PythonSCAD has an Align command which I need to wrap my mind around to see if it would be workable for positioning parts.
patience.
Wear hearing protection.
You more than likely will probably never need a table saw, jointer or a festool domino tool.
Tools I'd get first:
- japanese pull saws
- good chisel set
- a track saw... and a 2.4x1m.2m sheet of foam to cut on
- a palm router, learn how to make all your own jigs
- 12 irwin quick clamps. if you also prefer and can afford, 12 t-track f clamps.
- an impact dril, and a normal drill
- a decking drillbit
- a marking guage
- pencils
- a speed square
then
- make or get saw horses. and 2.4mx1.2m sheet of formply
100% avoid:
- 90 corner clamps.
instead get larger 1.2m irwin quick clamps or 3d print some corner protectors for a strap (assuming you're doing boxes)I have gotten just "ok" at CAD. I'm not as productive as I should be because I avoid it. Really looking forward to this.
The local version has a MCP server.
I'm doing mainly simple furniture and cabinets.
Isn’t this closer to being an architect or a designer now?
1. First several 3-d design iterations are throwaway reworked because I am constantly measuring the design, clearances and studs. And paper works best here instead of working with clankers which takes way longer.
2. Choosing the right kind of materials: thickness and type of plywood, type of hinges, panels, finished or unfinished. Optimize for simplicity of putting components together is most important.
3. Once the design settles the actual joinery needs to be as simple and efficient e.g. butt joints work well most of the time. A good jig for pocket holes is generally the most needed. You will be surprised how well a staple gun works for creating back panels of cupboards.
Its not that they are not useful, shopspec looks great(good job variodot), its just that I can get further and faster without them for most hobbyist pursuits. I invoke YAGNI most of the time because I constantly ask myself one question: "How can I make this be less time consuming?".
Coincidentally, I also seen a (private) demo using GPT-6 Astra designing wall panel joints based on some pre-determined rules, 2D lines specifying the layout, and later altering the design following a hand sketch (photo of a sketch). That looked promising too. It was claimed that spatial reasoning of Astra is elevated, can work well with 3D models and formats.
None of those are production (something to manufacture) level just yet. But if I had time I'd explore this manipulating CAD files by an agent route, opening those files in the usual tools I have at my disposal.
Talking/chatting about 3D is like dancing about architecture in my view, so I think this LLM use route is not sustainable straight, still, my above casual encounters suggest, for simple things, combined with manual adjustments, it may be alright.
RapidCAM.app is a parametric 2.5D CAD/CAM for CNC and laser with support for rotary axis. RapidCAM.app runs in the browser or as desktop apps. There is no login required. The app is free to use. Privacy is respected. rcamhub.com is the sharing site for .rcam files.
Within the App, you can define your project and then use the built in prompt generator. This creates a prompt with all the details AI could need. Once the AI replies with the file, you can paste it back into the app which will then verify the file for you.
Desktop versions of the app have MCP server built in so AI can control the app directly if that is a route you prefer.
I started RapidCAM for my own use as I found that the CNC was sitting idle simply because firing up a full CAD/CAM package took too long. The idea with RapidCAM is that you can quickly sketch out your job from as simple as you need to as advanced as you want going from unconstrained to full parametric/constrained as required.