@yacineMTB Did it pull camera model from grabcad and just cut it bounding box from cube ?
English
Justlovescience
141 posts
We're inviting hardware startups to try the next generation of CAD.
Prompt in → parametric 3D model out (STEP/STL/OBJ). No CAD experience needed.
Already working with OEM R&D teams. Now opening early access to startups building next-gen hardware.
Interested? Comment or DM.
English
@BrierRat SolidWorks is powerful, but I think its massive legacy code is causing it to fall behind. Fusion has a clean python api and already supports mcp, and I believe same is true for onshape. I'm not a big fan of fusion, but stuck with it for now.
English
@justlovescience Im not really in the fusion ecosystem. Solidworks for serious stuff. Shape3d for quick stuff. Claude CAD for the agent stuff
English
Claude CAD Fable, another one shot! This time reconstructing parametric from bare STEP
I gave a STEP file, a simple one-sentence instruction to steer build method, and it recreated the parametric model in build123d. Took about 10 minutes.
This is much easier for Claude than building from a layout, since Claude can just grep the STEP elements for the full geometry description.
Drawing included for reference. I built the part first from the drawing in Shapr3D, then exported the STEP and gave it to Claude to re-parameterize
István Csanády@istvan_csanady
@BrierRat How well it would work if you gave it the original brep model and it had to remodel it from scratch?
English
95% of this geometry can be analyzed with XFLR5 or even xfoil lmao. You can literally see vortices forming where the winglets make a sharp bend. You needto have no idea what you’re looking at to present these as something to be proud of
Sharp bends like this are guaranteed vortex generators. That geometry should have never hit CFD stage
English
I mean running permutations is pretty bomb but did it result in a design improvement or just generate a lot of heat and a large electrical bill?
nTop@nTopology
One nTop engineer set up a CFD pipeline, walked away, and came back to 12,000 completed simulations. 2,400 drone geometry variants, 5 angles of attack each, 280 GPUs fully saturated the entire run. No restarts, no manual fixes, no one watching it. @nTopology @CoreWeave
English
@BrierRat Hell yeah man, first vibe-cad I see that's not a junk! You should aim to teach fable how to make fits and try to 3D-print it, it will be a huge achievement.
English
@justlovescience Ask and receive! There are some quibbles. It is line to line, so fits weren't modeled (I never asked), but more design tuning with Claude can easily make this real from here. No intersections.
English
Fable CAD built me a crescent wrench from a picture!
Features full range of motion and jaws calculated to hold the full width of a max hex nut. Not perfect, not one-shot, but the model is pretty steerable. I feel confident I can steer it further for better accuracy
English
@10_X_eng First you need to check can it reach motor datasheet, grab mount holes sizing and spacing, then make mounts accordingly. If you can ask it "replace motor to this one" and it goes, grab the info from datasheet and make the right mount - it might be like... at least not bad
English
Ok, well the taste still sucks, but parts suck less than they did before.
English
@HuggingModels in nvfp4 it tends to stuck in endless thinking loop
English
Ever seen a 35B parameter MoE model squeezed into FP4 precision? NVIDIA's Qwen3.6-35B-A3B-NVFP4 is a text-generation beast that's 4x smaller in memory, yet keeps the smarts. This is efficiency redefined.
English
@ga0_lab @plastikmanfan69 this serrated washers are trash, use locktite 241
English
@plastikmanfan69 Yeah I forgot the names of those serrated washers again
English
fuck you *sheet metals your vice*
David Hansen 🇺🇸 🇳🇿@boxcardavid
These bolt-action PCB holders from Hakko are weird at first but great to use, hefty and very stable
English
@fraserpricee that was my first thought, second is : oh wow, thats sooo cool, I only have like 30 on my jetson thor. Would love to see concurrency numbers
English
@fraserpricee my experience with this model is it tends to overthink and a lot of time it stuck in endless thinking loop
English
@Ekaeoq @yacineMTB fusion will make gcode, for laser cutter you can go 3d printer's way - small raspberry handle cnc control, literally nothing stops us to have same software quality for cnc as we have for 3d printers
English
@yacineMTB Macos is already the better linux, most things I need don't even work on linux, but everything I need off of linux works perfectly on mac
The problem is laser cutting software that needs windows, but fuck it
English
An incredible amount of incredibly stupid comments defending Windows is just due to the fact that:
your average-bog-standard-Windows-pilled-slopfest- enshittified-ads-in-the-Start-Menu-cloud-telemetried-update-lottery-Electron-powered-OEM-junk-filled-plastic-laptop-fuckshit-trackpad-keyboard user cannot even begin to comprehend the supremacy that is MAC.
MacOS is the only reason why these 10+ year old devices still work, despite their piece of shit intel cpus and horrible thermal design.
Ekaeo@Ekaeoq
How is Windows this fucking shit, like honestly, how in the fuck does a trillion dollar company ship something this goddamn ass? I use one, a singular software app, that doesn't even make my macbook FLINCH, but completely destroys my 16GB, i5 8th gen win computer...... We're talking about an app that uses 0.6% cpu on my mac, an app that is not even optimised for silicon, but the same goddamn app uses at least 80% of my cpu on the windows laptop.... godforbid I touch anything, instantly shoots up to 99% and just freezes. HOW IS IT SOOO ASS, HOW IS A TRILLION DOLLAR COMPANY SHIPPING SUCH ASSSS THAT KEEPS GETTING WORSE, YEAR AFTER YEAR.... it's almost incredible honestly
English
@jun_song I still don't understand why there's no market of good (better than Qwen) models for 128GB. Deepseek V4 flash fits at Q2, which is a significant quality loss vs Q4. Minimax doesn't fit as well at Q4... Give us good models for 128gb max laptop performance
English
@dsjfnsdlkjfbsdl @BarathAnandan7 qwen 3.6 35b seems to easily fall to overreasoning
English
@BarathAnandan7 On my dgx spark, i usually run a qwen 3.6 35b model from sparkrun but when i have important work, i run qwen 122b with about 45 t per second. I sometimes run qwen 3.6 27b at 8q. I rarely use gemma. I serve all with vllm. i have lm studio and many models but i rarely use them
English
If you have a DGX Spark or any GPU at home,
What is the best local model you’ve been running on it?
The Gemma 4 26B A4B has been my favorite
Would love to try the suggestions out!!
English
@steipete fwiw im using that 52" dell and i freakin love it. no display issues but also am on linux not mac.
English
Got excited about the 52’ Dell, but it requires BetterDisplay hacking to get a usable resolution, but then I get weird flickers and also startup wake issues. [on M4 Max MBP]
So, Apple wins this round, 2xXDR it is.
dell.com/en-us/shop/del…
English