Show HN: GPT image editing, but for 3D models
I’m Zach one of the co-founders of Adam (https://www.adamcad.com). We're building AI-powered tools for CAD and 3D modeling [1].
We’ve recently been exploring a new way to bring GPT-style image editing directly into 3D model generation and are excited to showcase this in our web-app today. We’re calling it creative mode and are intrigued by the fun use cases this could create by making 3D generation more conversational!
For example you can put a prompt in such as “an elephant” then follow it up by “have it ride a skateboard” and it preserves the context, identity and maintains consistency with the previous model. We believe this lends itself better to an iterative design process when prototyping creative 3D assets or models for printing.
We’re offering everyone 10 free generations to start (ramping up soon!). Here’s a short video explaining how it works: https://www.loom.com/share/cf9ab91375374a4f93d6cc89619a043b
We’d also love you to try our parametric mode (free!) which uses LLMs to create a conversational interface for solid modeling as touched on in a recent HN thread [2]. We are leveraging the code generation capabilities of these models to generate OpenSCAD code (an open-source script based CAD) and are surfacing the variables as sliders the user can toggle to adjust their design. We hope this can give a glimpse into what it could be like to “vibe-CAD”. We will soon be releasing our results on Will Patrick's Text to CAD eval [3] and adding B-rep compatible export!
We’d love to hear what you think and where we should take this next :)
[1]https://x.com/zachdive/status/1882858765613228287
[2]https://news.ycombinator.com/item?id=43774990
[3]https://willpatrick.xyz/technology/2025/04/23/teaching-llms-...
One concept I explored was creating an interactive app where users can experiment with different material options—essentially a real-time configurator. There’s a great example here [1], where if you model an object as a .obj file (possibly similar to Adam’s parametric models), you can tweak its materials and colors dynamically. IKEA seems to have something similar in production for some of their products [2].
I experimented with Adam as well, and it did a surprisingly good job. The only catch: if you try to iterate too much, it tends to alter the form of the object. My ideal version of this would involve a professional photographer capturing high-resolution images of, say, a couch. Then I’d upload them into Adam, generate realistic renders with different fabrics or finishes, and download the final variants as high-quality images to use in catalogs or ecommerce.
[1] https://angon.me/experiments/6/
[2] https://www.ikea.com/gb/en/p/ektorp-2-seat-sofa-hakebo-grey-...
[3] https://app.adamcad.com/share/2f1e68ad-2cdd-4613-8fdc-fc33f2...
in that case wouldn't you bypass the 3d aspect altogether?
I asked for "something cute" and got a cute panda. I asked it to make its eyes bigger, and it did.
It wasn't a clean model by any means, but damn, pretty good considering I can probably fix this up faster than I could create it from 0.
Have you considered providing built-in tools for mesh decimation and UV unwrapping? I know it can be quickly done with meshlab, but I imagine not a lot of Adam users would even understand the need for decimation. Any possibility for also automating rigging?
I've been experimenting with image-to-image (and video-to-video) for basic texture projection, which I think shows promise:
https://bsky.app/profile/nickfisherau.bsky.social/post/3lqrl...
I just started diving into this today too:
https://github.com/YixunLiang/UniTEX
which works with in "volumetric space" (for lack of a better word), which I think makes a lot more sense.
we were thinking of rigging for the creative mode. we want to create more fun ways for our users to share their generations and animations could be a step towards that. would you be interested in that feature?
I think Mixamo nailed the autorigging years ago. Anything similar to that is good enough. If you guys want to go fancy, check out Cascadeur and what they are doing. For a cool skinning algo, you might want to check voxel heat difusing:
https://github.com/meshonline/Surface-Heat-Diffuse-Skinning
What about painting/iterating on the model, while respecting the UV constraints and boundaries of the original texture, is that on the roadmap?
The "creative" mode seems to be ok, but my main interest (the parametric) failed in my first test: generate a bottle.
But anyway, good job!
also would love to know what your use case is and why you are more interested in parametric vs creative!
Only the generated .scad would be great! I mean, I see to use cases for that that would be helpful for me:
1. Asking for a base model, download it as .scad and them improve it through OpenSCAD accordingly to my needs
2. Starting modeling in OpenSCAD, them ask the AI for some boring task (e.g. generate honeycomb patterns, hooks, hinges, and so on)
> also would love to know what your use case is and why you are more interested in parametric
Most of my use cases of 3D printing is for tools, household utilities, spare parts, etc. Because of that, my favorite tool is OpenSCAD, and I use it a lot.
But I reckon that it is sometimes really boring. Sometimes I need to spend a lot of time with trigonometry and other math tricks and less with modeling itself. For example, the aforementioned honeycomb patterns, I've spent some hours of my life playing with sines, cosines, apothems, etc while I think that it would be a job that an AI could do for me.
you can also use creative mode for functional single part objects. i've made cool cupholders, plant pots and desk ornaments with it
https://github.com/jhacksman/OpenSCAD-MCP-Server
To me "parametric" features seem like a nice to have vs a hard requirement. I just want to be able to talk at my computer and get back an STL to spec that I can print out
No comments yet
could you give me an example?
[0] https://doc.comsol.com/5.4/doc/com.comsol.help.heat/Introduc...
I have noticed that the tool does not do quite what I want it to. It reminds me of MidJourney a couple years ago. I might have to build a workflow where I design parts in Adam and arrange them in other tools. As details I supply for arrangement appear to being applied to individual parts.
Do you have any language that is getting better results than others?
https://app.adamcad.com/share/aa1a745c-a337-4ba6-a8c1-ca7298...
Where I figured I had poisoned the context somehow and was getting nowhere.
And here
https://app.adamcad.com/share/8b3dd548-511f-4b96-a1dd-ed34d4...
Where I got some traction but am working in the opposite direction. I am finding my language the barrier. I say "Apart" but how is it to know which way.
Will it respond to X/Y coordinates properly?
My next move is probably going to be feeding it some cleaned up diagrams when I have access to them.
Sadly, this appears to be a no go.
Or parametric structural optimization / topology optimization with parametric constraints?
Or, rather, if these are out of scope for your tool, then I'm wondering, like the other comment, how you'll slot into existing pipelines that do facilitate these techniques.
I'm not in the industry, I'm just curious.
For "creative" its a little different.
There are a lot of tools to convert figma into code, but is there a reverse? Say you have code already and you neglected figma. Any way to create the figma and iterate on it through an LLM?
I wonder if you think this is still true given how much better the foundation models are now.
But a lot of the data needed to train an AI is probably locked up in user company product lifecycle management systems. Autodesk is probably going to be out front in any AI wave of CAD with all the data they are accumulating in their Fusion 360 cloud. The main question for me is whether they will pull a Kodak and not develop it for fear of cannibalizing their current market, or not.
the future of cad generation will undoubtedly extend far beyond simple code-gen to generating human-readable features. this is where the data will be important!
Error Compiling Model — "Fix with AI?" — Okay — Hmmm.... only the top is truncated. (Seems to have done an "Optimized Tetrahedron"?)
Copilot + GPT-4 can generate build123d Python code; procedural CAD genai. It's spatially incoherent sometimes like there's not much RLHF for 3d in the coding LLM model, but the code examples are close to compiling and helpful. There's also an ocp-vscode extension to render build123d and cadquery CAD models in a tab in the vscode IDE: https://build123d.readthedocs.io/en/latest/external.html#ocp...
PartCAD is probably useful for your application as well.
Are there metadata standards for CAD parts, assemblies, and scenes? For example in PartCAD or in a BIM Building Information Model that catalogs part numbers and provenance for the boardroom hinge screws and maybe also 3D-printable or otherwise fab-able CAD models?
nething is an LLM with build123d output: https://nething.xyz/
FWIU it's possible to transform B-rep (boundary representation) to NURBS with a different tool.
B-rep: https://en.wikipedia.org/wiki/Boundary_representation
NURBS: https://en.wikipedia.org/wiki/Non-uniform_rational_B-spline
glTF 2.0 > Adoption of: https://en.wikipedia.org/wiki/GlTF#Adoption_of_glTF_2.0
PartCAD > Features > Generative AI: https://partcad.readthedocs.io/en/latest/features.html#gener... :
> Individual parts, assemblies and scenes can also can be exported into 3D model file formats, including: STEP, BREP, STL, 3MF, ThreeJS, OBJ, GLTF, IGES
Can you upload an existing model and start working from there?
Also, while I have you, could you please stop creating accounts for every few comments you post? We ban accounts that do that. This is also in the site guidelines.
You needn't use your real name, of course, but for HN to be a community, users need some identity for other users to relate to. Otherwise we may as well have no usernames and no community, and that would be a different kind of forum. https://hn.algolia.com/?sort=byDate&dateRange=all&type=comme...