Printing Gaussian Splats | Hacker News Reader

Printing Gaussian Splats

patreon.com

375 pointsilnmtlbnm46 comments

Incredible. I may be the only one in the dark, but until this moment I had no idea 3D printing at this high a fidelity was possible. It looks like a real bee.

Last I knew, the best 3D prints still looked like hardened play dough

I do a lot of 3D printing and I had no idea either. I did some searching and this printer is as big as a work bench and likely costs around $200,000.

It's different kinds of printers. That's a resin printer - pretty high-end one. I'm not getting that result on my Bambu P1S.

But how does resin do colors?

One approach is to just print CMYK resin like an inkjet printer (and then cure it with UV).

Do that hundreds or thousands of times and you eventually get Z height.

Look up the EufyMake E1 for a consumer/prosumer version.

The EufiMake isn't really a 3D printer though. It's more a normal printer that can do an embossed/textured print.

UV Inkjet.

HeyGears is releasing a prosumer full color UV inkjet resin printer this year for < $2K: https://store.heygears.com/products/heygears-g1-direct

Well, to actually 3D print looks like it’s $3300, but still that’s really incredible! Seems substantially less messy than the resin printers I played around with a few years ago.

Heygears don't seem to have the best track record https://www.reddit.com/r/3Dprinting/comments/1bu3mv1/avoid_t...

Calling it a resin printer is like calling a FDM printer and injection molding machines in the same category, both can melt ABS but the way they work and capabilities are completely different.

Same thing here hardly anything common with hobbyist resin printers beside using some kind of UV curable resin. And as with other 3d printing technologies Stratasys is decade ahead in terms of research and commercialization sitting on all the relevant patents and selling expensive machines (sometimes as a result of acquisition).

Once the patents ran out maybe there will be more advancements and general availability. Although I expect much longer delay compared to FDM and SLA/DLP 3d printers. Inkjet printing on paper is already complicated and finicky enough, It's not something a hobbyist can make from scratch in a garage. Add a resin which will by design solidify when exposed to light potential destroying the inkjet nozzles, and doesn't necessarily behave as regular ink when attempting to spray it through inkjet head and you get the need for some serious investment to recreate the technology even with patents expired. The recent hobbyist 2d UV printers are step in this direction, but commercial/industrial UV printers have existed for quite a while. To me this suggests there is additional gap in patents/technological challenges between textured 2d UV printers, and full 3d UV inkjet printing.

Same. This is insane, I had no idea!

There's nothing I personally want to bring, but these would make AMAZING gifts, cool things for your desk/bookcase, etc.

There have got to be so many interesting, educational, and cultural objects you could print like this, and the fact you can "blow up" an object like a insect is even cooler.

Depending on the price, this feels like something that could take off in a big way.

There s a whole spectrum between hardened play dough and the bee in TFA. A $200 printer with a 0.2 mm nozzle and proper setup (ironing calibration etc.) is already capable of making 3D prototype prints with details that look professional (e.g. highly legible tiny fonts).

An example would be the multicolor articulated dragons that are flooding flea markets and garage sales around the world : when printed properly they are highly detailed and looked mass produced. Unsuspecting parents buy those for their kids and have zero idea these were 3D printed on $200 printers.

The 3D printing world progressed a huge lot in a few years, which is what prompted me to buy one. It progressed so much it s basically solved.

Once in a while I get inspired by a post on HN, but I know I don't have the energy to follow it through. Here's my idea for anyone who thinks they can do something with it:

Imagine you are holding in your hand a small glassy sphere. Initially it looks like random dots inside, but as you play with it in your hands, you get glimpses of images of loved ones.

To implement this, select a bunch of photos of people, generate pseudo-3d models of their faces with your AI of choice. Position them around a sphere in a 3D modelling software, and generate a whole bunch of renderings all around the scene. Now, feed those renderings into gaussian splat software.

The resulting 3D gaussian splat should include relatively high fidelity when the viewer is aligned with the supplied iamges, but more stochastic nonsense when misaligned. This should contribute to the feeling of being able to 'find' images while rotating the sphere around.

Not exactly what you describe but this project [0] did it with cubes.

The cube is filled with splats. Each face reveals a different picture when viewed from a perpendicular angle

[0] https://www.3dpoint.art/

if you allowed for some slippage in precision, how many images could you fit in that are recognizable from various angles, with ~acceptable~ resolution? Sounds fun.

This is the splat https://superspl.at/scene/3ae6a716

Overview of the printer used: https://youtu.be/T6sxVtViX-k

Wow, truly amazing that you can print a high fidelity Snickers bar.

Very cool product! And to think, in one of the many prior gaussian splatting threads someone declared there was no way anyone could build a business around the technique.

To be fair, if they first vocalize it they could just use Sparse Voxels which is my favorite differentiable rendering technique https://github.com/NVlabs/svraster

I find the workflow of fitting Gaussian splats and then immediately turning them into voxels to be rather surprising, too. I can imagine some performance benefits at large scale, or maybe a reduced tendency to generate certain types of artifacts, but going straight to voxels seems more straightforward.

What is the 3d printing technique being used here? I can’t intuitively recognize it.

Resin, on one of these big expensive Stratasys machines: https://www.stratasys.com/en/3d-printers/printer-catalog/pol...

so they buy and operate the machine, the customer brings the splat. is this just a print lab?

Looks like it. Plus it looks like they’re closer to a traditional screen sprinting shop in that they just take a file and will work with you manually tweak and adjust things to look good, rather than just printing a file straight away

I remember seeing a 3d printer that was essentially a 2d printer which printed the surface and cut the outline and then laminated the sheets together which depending on the paper used would get you a block of something between mdf and plywood with high surface detail imagery.

I would imagine much the same approach could be done by laminating clear plastic sheets if you can maintain the transparencey without bubbles. It would get you modern colour printer resolution in two dimensions and sheet thickness in the other.

It wouldn't surprise me if some smart cookie could make a resin printer with a resin that sets in a state reflecting different wavelengths depending on how you zap it. That's a problem left for the reader.

You could easily release pigment into resin just before it gets hardened, but getting the right pigment to the right place would be hard, A print head zapping back and forth inside the liquid doesn't sound like it would be viable.

Printing in resin bottom to top part could allow a colour print head to fly over the surface printing a pigment layer then squirting the next layer of resin on top, zap and repeat.

How much are the prints approximately? Can't see a price anywhere.

Not sure, but this will be cheap soon and invading trinket shops everywhere for sure!

Sure beats flexy dragons, so I'm all for it.

You know exactly what I was hinting at haha

$143 according to a comment on the page.

Trying a bunch of sizes in their studio tool (all cubes to make it easier):

5mm x 5mm x 5mm (smallest I can select): $32.58

10mm x 10mm x 10mm: $36.58

20mm x 20mm x 20mm: $82.86

30mm x 30mm x 30mm: $143.60

50mm x 50mm x 50mm: $509.14

100mm x 100mm x 100mm: $3,028.58

150mm x 150mm x 150mm: $9,618.86

195mm x 195mm x 195mm: $20,641.32

(it says it's an estimate excluding tax and shipping)

It does not include shipping. I paid an extra $30 for shipping a 40mm cube. It’s the only reason I didn’t comment initially in support of an otherwise cool service

Interesting how they voxelate the splat to print it. I am sure there are more efficient algorithms that allow you to make a mesh which then gets sliced before printing

I am wondering if it would be possible to print a splat similarly as it is shown on the screen, so not voxelized, but like a hologram with overlapping transparent gaussians. It looks like the model from the outside at the given angle, but the inside is nothing like model itself.

Resin printers like this use UV pixels to cure the material instead of extrusion nozzles to trace the perimeter. Voxels are the natural format for resin.

But, if you made a mesh you wouldn’t be able to represent fuzzy objects. That’s one of the major advantages of both splats and this printing technique.

This is insanely cool. My wallet is ready.

This is genuinely fucking cool. I don't think I've ever seen a 3d printing implementation like this before. I'm not a big fan of splats as a modeling method, but for something like this it's fantastic.

I can already see myself hoarding all the prints

Definitely ready to try this. Some scifi stuff

Why it's called Gaussian?

The original paper that kicked off the movement a few years ago used Gaussian blobs because gaussians are differentiable and the tech to analyze a scene requires the tech used for rendering to be differentiable.

https://learnopencv.com/3d-gaussian-splatting/

Absolutely incredible

I want buy one.

Fantastic work!

Reply on news.ycombinator.com