Re: Weekend Update 5/10/26

[sirenwerks]

Another experiment using ai to produce a 3d model from a photo.
Ai isn’t the end-all for developing models but it’s a good start on one-offs like Milwaukee’s 1954 Ford rail detector.

(Attachment Link)

Here’s my finished model.

(Attachment Link)

At the risk of being 'that guy' and criticizing an amazing model, AI switched the rooftop lights around. I only noticed because your models are frequently lit and I was drawn to the lights like a moth.

[dem34]

At the risk of being 'that guy' and criticizing an amazing model, AI switched the rooftop lights around. I only noticed because your models are frequently lit and I was drawn to the lights like a moth.

Which then drew my eye to the massive increase in ribs.

[Jim Starbuck]

Which then drew my eye to the massive increase in ribs.

If achieving prototype fidelity is the goal ai isn’t the avenue to pursue. However in creating an obscure vehicle that will populate a square inch on the layout and likely go unnoticed I chose to employ the wisdom of the ancient Stoic philosopher Mediocritese who wrote “Meh…Good enough”. 

[Chris333]

Yeah I can pick out all the problems with it. But I still can't believe it did that good. It is very impressive. Wait till they know what their doing.

[nkalanaga]

That's actually the kind of thing today's AI should be good at, once it's perfected.  No imagination or creativity needed, just analysis and drawing.  As long as there's enough data to do the calculations, a computer shouldn't have any problems creating a 3D print file.

[Maletrain]

We already have the laser scan prints of Ed (and anybody else who is willing to some day end up with a robot that looks, sounds and acts like him/her).

And there is also already at least one computer program that can do phototomography with a pair of regular cameras spaced about a meter apart, using them to take photos from a bunch of angles.

So, the limitation will probably be for historical structures where there are only limited views, maybe taken years apart.  That is going to require some critical thinking about how to deal with modifications and get details from other similar structures.  So, actual modeling will probably  endure for those of us who don't do "modern" (whatever that becomes).

[John]

If achieving prototype fidelity is the goal ai isn’t the avenue to pursue. However in creating an obscure vehicle that will populate a square inch on the layout and likely go unnoticed I chose to employ the wisdom of the ancient Stoic philosopher Mediocritese who wrote “Meh…Good enough”.

I agree with Jim.  I often go down the rabbit hole when I do 3d models… for example Dave’s Huntington station was at least 4 or 5 tries to get it “close” to where I was satisfied, but a vehicle like this from AI is a good starting point.  If the ribs or minor details like lights are a show stopper for you, a little blender or other 3d tool can fix those imperfections easily enough.  What’s cool is only Jim has one of these right now.


Sometimes you just have to accept that good enough is a standard.  Paralysis by analysis or “what if “ brings projects to itheir knees.   

[davefoxx]

Sometimes you just have to accept that good enough is a standard.  Paralysis by analysis or “what if “ brings projects to itheir knees.

Agreed.  As the old saying goes, perfection is the enemy of good.

DFF

[Jim Starbuck]

Agreed.  As the old saying goes, perfection is the enemy of good.

DFF

It also depends on the model and how far you want to take it.
For instance the 90 ton Whitcomb oil electric is on it’s fourth round of refinement and third chassis design. Another element that comes into play for me is that I’m new to 3d modeling and apply techniques as I learn the various software.
Quick cheap and easy projects like the van keep me motivated to tackle more sophisticated models.
Sometimes you draw with pen and ink and sometimes you color with crayons.

[jjb62556]

I like it...Period!

[randgust]

I've had multiple generations of the 'same model' since I started N in '72.    Started with Trix F-units, evolved up into two generations of Kato.   GP7's, started with Rapido, now Atlas second generation.   RSD15's, just retired Model Power now gone to BLI.   Sawed up shorty beer can tanks to make decent 50' modern tanks, now Atlas.  On and on.   Get as close as you can for now, have no qualms about replacing when something better 'finally' comes along.  Or it may never.  But don't wait for it.

The quality of RP printing is now just staggering to me.   The Shapeways era was rough, not anymore.   The surface finish I'm seeing now is astounding.  The only part I have to chuckle with is watching the ongoing struggle with handrails and grabs when it's like wire was never invented.   Not everything deserves printing; fine wire and wood grain included.

The true irony is in the 'end of conventional modelbuilding' we now need paint and decals on blank RP prints more than ever.  I can print some decals, but the loss of Testors, Floquil, Pactra, makes all those little jars priceless.   For fun, go on to Ebay and price out a single jar of Floquil (original) thinner.  The empty cans now are selling.

[Dave V]

Yeah, I've been saying that...that if 3D printing is the future of the hobby, we need our railroad colors back!

I've been able to come close in a lot of cases with Vallejo paints designed for military modelers (for whom paints are even more vital).

[Scottl]

Paint mixing is not rocket science.  While it is nice to buy some pre-determined colour off the shelf, "scratching" a suitable mix is straightforward. The military and miniatures hobbies are keeping us well supplied with quality paint brands and novel new products.

And while I acknowledge the limits of the printing technologies available, I think printing wood grain and fine railing profiles is both feasible and in many cases, visually superior to wire methods.  Sure, I can futz with wire, but it will look lousy compared to straight off the printer.  In many cases, an entire, complex model can come off the printer in one pass.   

AI model creation is another story.  Lots of potential and I've had good success generating models that are good enough, in some cases excellent. But rivet-counters still need to look elsewhere, for now.

[davefoxx]

Paint mixing is not rocket science.  While it is nice to buy some pre-determined colour off the shelf, "scratching" a suitable mix is straightforward.

Not for us colorblind folks.

DFF

[nkalanaga]

Or for anyone who has to worry about chemicals, whether odor or toxins.  I can't spray indoors, and have to be careful what I brush paint inside.  Using an airbrush outdoors would be difficult most of the year.  Almost all of my painting is done with spray cans - run outside, paint, run back in.

The big-box-store craft paints work fine for weathering, but don't do well for base coats.