3D print people figures

[Sam95]

Dear all,

I have been busy on other subject for the past month but I am back to model railway. As my station is empty, I searched an easy and cheap way to have figures on the platform.
I have a Sidewinder X1 printer so I tried to print with this printer to see the result.

I tried Makehuman and blender but I had some errors in the print (holes in the face due to the presence of eyes and mouth cavity in the makehuman model) I am still looking how to correct that.

But during my time spent on youtube I found a video talking about an AI to generate 3D models. So I tried and the result are not bad at all !!

So the first thing to do is to register on :
https://www.meshy.ai/

You can go for a free account, because you have 200 token each month so you can generate few models. Actually it costs 10 token to generate a model.

Once connected you have to go on "text to 3D print"

Write a prompt to describe what you want and clic generate. Few seconds later the AI will propose you 4 designs. Here is an example :


mettre une photo sur une image


You can choose either one of the 4 desing and check if it is ok for you, then you can download it directly in STL for free.

After that I imported the STL into Cura slicer, scale it down to fit G scale (1:22,5); and that's it !!

To print it on my sidewinder X1 with elegoo grey PLA I use a 0,4mm nozzle with 0,1mm for the layer height, other parameters are for the PLA standard to avoid stringing and other problems.

Here is the result juste after printing and after support removal :
















The quality is very nice even if it is not a resin printer. It took about 3 hours to print this model, but I have plenty of time for that.

The next step will be the final clean up and the painting..... I will probably ask to my wife to do it !

I hope this will give you some ideas, and after we can share our STL

[philipy]

That looks most impressive... I did wonder if AI would ever get useful for us. I've tried and got nowhere, personally!

Please keep us uptodate as you progress... even if your wife does do the painting.

EDITED to add that I've just had a quick play online and it does work pretty well. The trick seems to be in the way you write the "prompt" and I'm not very good at that sort of thing but I guess some practice will help. I also notice that there is a plug-in for Blender which I presume will make posing the figure much easier.

[ge_rik]

Sounds and looks really interesting. Especially if the AI commands make the process easier.
BTW. What was the text prompt you used to create that figure. Hopefully, we don't have to brush up our schoolboy French.....

Rik

[GAP]

That looks most impressive... I did wonder if AI would ever get useful for us. I've tried and got nowhere, personally!

Please keep us up to date as you progress... even if your wife does do the painting.

EDITED to add that I've just had a quick play online and it does work pretty well. The trick seems to be in the way you write the "prompt" and I'm not very good at that sort of thing but I guess some practice will help. I also notice that there is a plug-in for Blender which I presume will make posing the figure much easier.

I'm using AI to write a computer program that will let me control my remote points and signals plus display my train locations on a screen via wifi .
The correct prompts/instructions is very important and that is the mentally challenging part.
When I am at the figure printing stage I'm going to give this a go.

[philipy]

What was the text prompt you used to create that figure. Hopefully, we don't have to brush up our schoolboy French.....

Rik,
I've been having a deeper look at this and think I'm making some progress. There are quite a few Youtube etc so-called guides which almost without exception are aiming at fantasy D&D type non-humans. None that I've found tell you how to generate a realistic 19xx's loco driver!!

No you don't need French can do it in English.

Soooo.... I've tried to describe a fairly basic, generic workman type figure, just to see what it comes back with.

My "prompt" ( which is what it calls it ) is as follows:
Male human figure 5ft 6inches tall standing with right leg slightly forward of left leg. Slightly stooped forward pose with right arm raised to shoulder height. Somewhat worn workmans trousers, shirt with rolled up sleeves and laced shoes but no hat. hair short and balding

The figure generated from that is:

Screenshot 2024-12-07 06.14.38(2).png (141.39 KiB) Viewed 100813 times

So, it appears to be unable to tell right from left, obviously I didn't tell it to point upwards and it seems to have ignored the short hair instruction and just gone for bald all over!

There does seem to be a lot of potential in this though, but it will depend on ones ability to describe the desired end result in great detail.

Edited to add that I've just discovered the "Texture" button, which auto colours it and shows a lot more details than the basic grey version. The hair is there but of course I didn't tell it what colour, so it is seems to have done it very pale - my fault.

Screenshot 2024-12-07 06.35.39.png (154.28 KiB) Viewed 100811 times

[philipy]

As well as describing a person to generate a model, it also has the ability to generate a 3D figure from a 2D photo.

I've been playing and, from L - R... a lady in a sparkly dress... the basic generated model... the same thing textured... a rear view of the textured version, which is obviously completely synthesized. It has knocked a few years off her apparent age and for some reason has adjusted the position of her arms and handbag, but that is a minor point.

Potentially this could enable specific real engine drivers, or whatever, to be generated wearing their correct clothes or holding tools or whatever, and then 3d printed!

082a.jpg (234.13 KiB) Viewed 100808 times

[ge_rik]

Really interesting. The poses so far all look fairly upright and static, I wonder how it copes with a figure in a more active pose. I'm away from home at present but will start experimenting when I return home.

Presumably, being AI, it will learn from experience and so the more it does, the better it will become.

Rik

[Scrat]

As well as describing a person to generate a model, it also has the ability to generate a 3D figure from a 2D photo.

I have to try that.

Background : A friend (M J Price from Llanelli) built some magnificent 5" gauge locos. As he sold one unfinished project to me I wanted him as driver in his Reichsbahn uniform.
Sadly, the company went bust and all I got was promises and this set of pictures. As John died a few years ago this is all I have to start with.
Will give it a try.

[philipy]

Really interesting. The poses so far all look fairly upright and static, I wonder how it copes with a figure in a more active pose. I'm away from home at present but will start experimenting when I return home.

Presumably, being AI, it will learn from experience and so the more it does, the better it will become.

Rik

Just had a quick look and can't find it now, and I haven't tried it, but I'm sure that when I was exploring I found various pose options such as running, T-shape, etc.

[ge_rik]

Just had a quick play on my phone. The interface isn't really suited to a small screen but here's what it produced with the prompt "vintage workman swinging axe to chop a tree"

Screenshot_2024-12-07-09-14-53-018_Edited_Edited_Edited_Edited.jpg (51.88 KiB) Viewed 100794 times

Rik

[philipy]

Seeing your axeman , I had an idea and tried an experiment.
I put in "Festiniog and Blaenau 4 wheel railway coach".

I got 4 very different offerings, none of which more than remotely resembled a prototype, which was no real surprise, I didn't really expect it to know a pretty obscure and short lived Welsh line, but this is not a bad stab at a generic 4-wh coach.

Screenshot 2024-12-07 11.56.22.png (183.06 KiB) Viewed 100778 times

[Phil.P]

The workman figure (with the raised arm) looks very Clockwork Orange, to me.

But it does look to be a very interesting development.
If it is flattering when converting photos, I might even be able to end up with a mini SW!

Phil.P

[Sam95]

I am happy to see that you are playing with my discover !!

Indeed the prompt text seems not to be very well understood sometimes. Yesterday I asked for a woman giving hand to her son.... Finally I had the woman and the boy standing side by side but they were not giving hands.

I also tried to copy and past a prompt from an available model, and the results were totaly different...

According to the bellow video it is possible to change the pose of a meshy model into Blender :


If someone wants to test becaus I am not at home this week end.

I think you can achieve much more different models if you use Makehuman and Blender but it is a more complexe process.

My initial goal was also to see if a FDM printer can print nice model at our scale.

[GAP]

I am happy to see that you are playing with my discover !!

Indeed the prompt text seems not to be very well understood sometimes. Yesterday I asked for a woman giving hand to her son.... Finally I had the woman and the boy standing side by side but they were not giving hands.

I also tried to copy and past a prompt from an available model, and the results were totaly different...

According to the bellow video it is possible to change the pose of a meshy model into Blender :


If someone wants to test becaus I am not at home this week end.

I think you can achieve much more different models if you use Makehuman and Blender but it is a more complexe process.

My initial goal was also to see if a FDM printer can print nice model at our scale.

Seems to have done a fairly good job

[drewzero1]

I generally dislike AI for a multitude of reasons, but I have to admit that in some applications (such as Graeme's, for instance) it can be actually useful. I've been wanting some more printable figures and after my own dismal attempts I reluctantly gave the AI tool a try.

Screenshot_20241207-130930.jpg (97.61 KiB) Viewed 101095 times

I clearly need to work on my prompt writing skills, I did not expect I would have to specify that the man be holding the briefcase!

[drewzero1]

I also had a go at the photo-based generation, using an old family photo. I thought it did a good job capturing the proportions and pose and to some extent the clothing. The AI seemed a bit confused what he was doing with his hand, but to be honest, so am I.

rfSHEhBWmFoDUAAAAASUVORK5CYII=.png (972.23 KiB) Viewed 101094 times

Printed on Ender 3 V2.

[philipy]

That looks pretty good Drew. I would say the variations from the original are about on par for the way it modified my photo yesterday.
I doubt that there is much that can be done about that other than taking it into Blender and manually correcting things, and that is something I don't even want to contemplate doing!

[Paul_in_Ricky]

I doubt that there is much that can be done about that other than taking it into Blender and manually correcting things, and that is something I don't even want to contemplate doing!

I'm baffled by Blender, but there a couple of good step by step tutorials around for taking STL figures and reposing them in Blender I've managed to follow.

[Paul_in_Ricky]

Anyone know what the copyright situation is with these ?

<Later> Seems to be copyright free on the 'free' plan, but owner's copyright for the paid subscriptions.

[Paul_in_Ricky]

Just registered and had a quick go.
Remarkably decent looking figure that's close to the text description.
What's not so good is some files have multiple holes in them, so won't print. It seems there'll be some repair work needed before I can actually print one and a learning curve on how to do that :-(