3D print people figures

[Paul_in_Ricky]

I am amazed at how it manages to infer (or should it be extrapolate?) how the rear of the figure would appear.

But it doesn't always get it right

hitme3D and prompt.jpg (117.33 KiB) Viewed 990 times

[Durley]

Impressive model from such an apparently poor image.

Meshy also added a 3d version of the watermark "Alamy"

I wonder how long it will be before they detect copyright information and refuse to process it ?

On the subject of copyright, I did do some digging into copyright law out of interest. In the UK, copyright on photos is generally owned by the photographer for their life plus 70 years after death for images taken after 1989 and their life plus 50 years for images taken before 1989. After that point the images become public domain. Rules in other countries will be different.

That means I think we are on relatively safe grounds using images of British subjects from around Edwardian periods and earlier, but more recent images could still be under copyright and should seek the copyright holder’s permission to use the image. Alamy and similar sites sell images for use under license as dictated by their contract of use, which will impose additional terms.

I suspect there is not likely to be issues using images for personal use but could be a potential consideration if generating models to sell commercially.

I did find Google Gemini refused to work on a Victorian era image I had downloaded from a Facebook ‘History of Southampton’ page, so there must be some inbuilt digital protections being applied somewhere.

[Paul_in_Ricky]

As a photographer who has work sold by Alamy I'm aware of general copyright issues.

The photo of the Victorian mill worker is an interesting case as Alamy are distributing it for commercial use. Copyright can be passed on (assigned) to non-originators, so just because a work is old doesn't automatically mean it's out of copyright.
For example the photo of the Plynlimon and Hafan tramway posted here is almost certainly from Ted Wade's book that's still within copyright. I can't remember where that photo originated, maybe in the public domain, used with consent and maybe used for a small fee or owned by Ted but without copyright assigned to him. It's a long time since I helped him with that research.

How copyright might be interpreted in the case of a photograph heavily modified by AI and then used to generate a 3D model is something the lawyers will scratch their heads over for some time. It would be different if the modified photograph itself was being used for commercial purposes, but once it turns into a 3D mesh and there's significant amounts of new content, that muddies the waters a lot.
Worth considering that the AI companies claim copyright for non-subscribers (ie only subscribers can use the models commercially). Quite how that can work if the prompt's copyright doesn't belong to the subscriber will be another muddy issue, although I suspect buried in the T&Cs will be a clause saying prompts must be the user's own work.

Pragmatically for the case of model railway figures there's so little money involved it would never be worth taking legal action and the worst you'd be likely to suffer is getting a 'cease and desist' letter from a solicitor.

[Durley]

Thanks for the detail Paul, good to get some insight from someone who knows what they are talking about!

[Trevor Thompson]

Good as it is for generic type pictures of figures, I've been a bit disappointed with the way that Meshy distorts photos of real people. However, by accident it has just nearly blown my mind!

I came across a picture on Alamy of a girl operating a Victorian cotton mill. Thinking of Trevor wanting female workers, i downloaded the picture and roughly edited her out and bunged it into Gemini. For some reason, it deleted most of the background but left part of the loom in. Not expecting anything usable, I loaded it into Meshy anyway, and Wow!

Screenshot 2025-10-03 21.38.10.png



2nd optimised.gif

I wasn't expecting anything usable, but merely some garbage to edit out. i doubt it would print well though, because there is just too much that is too small to support. It does potentially open up possibilities though. F'rinstance, how about a signal man and all of his levers, printed as one unit?

That is brilliant. It is clearly going in the right direction! That female figure is very usable - without the loom - but of course the loom is also interesting. Weaving was a cottage industry in victorian times so appropriate even in the Slate mining areas. I can just see a row of weavers cottages with windows in the roof, with the looms and weavers visible on the top storey - alongside the railway. That has set me off on another tangent!

Trevor

[Scrat]

WOW. Amazing results.
I will try that on John´s pictures.
Maybe I´ll get something useful.

[Paul_in_Ricky]

I’ve just found and resolved a curious side issue to these high resolution STLs generated by AI (Tripo, Meshy etc).

Trying to get a figure to an exact pose via AI seems still rather too hit and miss for poses that require some precision like loco crew. To get a more exact position for one of my STLs I tried using philipy’s excellent tutorial ( viewtopic.php?t=14754 ) on re-posing figures in Blender, something I’ve followed before without problems.

However at the stage of attaching the skeleton to the AI mesh I persistently got an error
“bone heat weighting failed to find solution for one or more bones”
This seems a common problem from a quick Google, but all of the suggested solutions that mainly involved reducing mesh resolution failed to work for me.
It seems the problem lay with the figure’s lack of appropriate dimensions. Loading the figure into a program that can scale and then re-save the figure at the size you need seems to solve the problem. I’m sure Blender ought to be able to do this itself, but I haven’t found a way yet and it’s sizing options make little sense to me.
I ended up using my printer’s slicer software, Bambu Studio. When the AI mesh is loaded the software first complains that the “model is too small and may be in meters or inches. do you want to scale millimeters ?” (sic) That loaded a tiny model with 2m+ triangles and 25mm high. Rescaling by 255% brings it to 16mm scale (it’s a sitting figure, so the exact rescaling will vary). It’s then possible to re-export the figure as a STL from Bambu Studio at the final size and take it into Blender and rig it. Worth noting that, I assume, because of the level of detail attaching the armature to the mesh takes many minutes even on a fast PC, patience was needed.

Worth noting again that this is only really possible if the limbs are separate from the body when built by the AI, so prompts need to be thoughtfully created to generate a pose that can be later modified.

[Durley]

That’s something I’ve still to try Paul but sounds promising. Did you start with a T pose for the figure to be repositioned in Blender, or is that not necessary? Can Blender reshape hands or do you need to have the hand shape pre-determined (open hand, gripping, etc.)?

Interestingly and contrary to what you are finding, all the figures I have downloaded as STLs from Meshy are huge, the default appears to be 10m tall unless I resize the model when downloading.

[philipy]

I always choose the resize option in Meshy before downloading. The default is metres and 1, so you end up with a 1metre tall figure which Cura and Chitubox don't like!
I always change it to mm and 85 or 90 and then do a final resize in the slicer. I haven't found a way to reset the default in Meshy.

AFAIK, you do need to get a T-pose from Gemini to get Meshy to produce a Blender rig-able figure, at least that's my experience so far. So I've had Gemini produce the figure I want and then when happy I've told it to convert to T-pose but not change anything else. That seems to work - watch this space!

In terms of hands, it depends what you want to do, you can put as many bones in the hands/fingers as you want in Blender ( or none at all if you don't need bendy fingers)

I had the same experience as Paul, with the skeleton not being fixed in the body. I was in the middle of trying again this morning when I got interupted and lost what i had produced!

[Paul_in_Ricky]

That’s something I’ve still to try Paul but sounds promising. Did you start with a T pose for the figure to be repositioned in Blender, or is that not necessary? Can Blender reshape hands or do you need to have the hand shape pre-determined (open hand, gripping, etc.)?

I've only had a short play with this. My impression is that we should get the pose very close to what we want, with the constraint of keeping limbs apart from the body. I think if you try to make drastic pose changes controlling mesh distortion becomes difficult, if not impossible. So it's probably best thought of as a fine adjustment, rather than a complete blank page for posing.

Interestingly and contrary to what you are finding, all the figures I have downloaded as STLs from Meshy are huge, the default appears to be 10m tall unless I resize the model when downloading.

Indeed that's what some software sees them as. However I think there's some curiosity about the size data that causes some software not recognise that or can't handle it (as seems to be the case with Bambu Studio) and needs resizing. Blender reports the height as about 5.7m for the sitting man I've been working on, but the output STL is just 65mm high, I don't pretend to understand that.

[philipy]

I had the same experience as Paul, with the skeleton not being fixed in the body. I was in the middle of trying again this morning when I got interupted and lost what i had produced!

OK, I tried again, and the bonifying does work.This is a Gemini created image based on a scan from a figure in a book. Once happy I then told it to make into a T-pose, then processed through Meshy and then rigged in Blender and posed! The posing is quick and crude just to prove to myself that it does work. I've got no idea why I had a problem before.

The only thing I'm not happy about is that I wanted him to have an open neck shirt and neckerchief, but Gemini wouldn't let him have a bare chest and Meshy insists on giving him a collar and tie! I can see some surgery coming.

Posed - Before & After.jpg (173.73 KiB) Viewed 498 times

[Durley]

That’s great Phil, it demonstrates fine tuning of the figure’s pose is achievable so we can tweak e.g. loco crew to fit a specific loco cab, etc. I now need to get on and learn how to use Blender!

[philipy]

... fine tuning of the figure’s pose is achievable so we can tweak e.g. loco crew to fit a specific loco cab, etc. I now need to get on and learn how to use Blender!

Yes, it does work quite well for that. Back in 2019 I had a project to build a weird little contractors loco (viewtopic.php?t=13763&hilit=chingford&start=100) and I made a crew for it by creating figures in MakeHuman and then posing them in Blender ( the link that Paul posted earlier). It means that you can create realistic poses that would never be available in any other way,i.e the fireman hanging out of the side as this loco reverses, and although you can't see it, the driver is actually holding the regulator. I have made a crew for my F&B 0-4-2 but now that we have the Gemini/Meshy option I'm going to redo them because it's more realistic than Makehuman.

DSC_0001[1].jpg (891.92 KiB) Viewed 483 times

However, unless you are a total masochist, "learning" Blender is impossible and not even worth thinking about. All you need are a few basic functions/commands and then follow my tutorial ( he says modestly). That will get you to be able to pose stl's

[Durley]

However, unless you are a total masochist, "learning" Blender is impossible and not even worth thinking about. All you need are a few basic functions/commands and then follow my tutorial ( he says modestly). That will get you to be able to pose stl's

Duly noted!!! I’ve previously kept clear of Blender thinking of it as more of an artistry tool and therefore outside of my engineering comfort zone.

[ge_rik]

I had the same experience as Paul, with the skeleton not being fixed in the body. I was in the middle of trying again this morning when I got interupted and lost what i had produced!

OK, I tried again, and the bonifying does work.This is a Gemini created image based on a scan from a figure in a book. Once happy I then told it to make into a T-pose, then processed through Meshy and then rigged in Blender and posed! The posing is quick and crude just to prove to myself that it does work. I've got no idea why I had a problem before.

The only thing I'm not happy about is that I wanted him to have an open neck shirt and neckerchief, but Gemini wouldn't let him have a bare chest and Meshy insists on giving him a collar and tie! I can see some surgery coming.

Posed - Before & After.jpg

That's excellent. I must give that a try when I get back home.

Rik

[Paul_in_Ricky]

OK, I tried again, and the bonifying does work...... The posing is quick and crude just to prove to myself that it does work. I've got no idea why I had a problem before.

Good that you've sorted out the boning problem. Although that illustration could be a good example of why you shouldn't try to move limbs too far as the elbow and shoulder joints look rather distorted.
It might be possible to get round this by manually weighting the attachment of the armature to the mesh. I haven't tried this myself yet, but watched a tutorial on it and the technique looked to have potential.

, but Gemini wouldn't let him have a bare chest and Meshy insists on giving him a collar and tie! I can see some surgery coming.

Gemini is weird sometimes. It will get in a loop and just repeat the same output despite prompts asking it to change. Saving the output of a 'stuck' chat and putting it into a completely new chat sometimes resolves this. It's as if you're just talking to someone completely different that understands what you want.
But then sometimes it just never seems capable of delivering what you ask for, maybe we aren't asking for things with words and descriptions it understands ? maybe lack of training ? it can be very frustrating.

[philipy]

Paul, You're right, but the lumpy shoulders were due to the fact that I (Gemini) originally created a vertically standing figure, then told it to change it to an T-pose for me to export. I think it just moved the arms without considering that that would affect the body as well. It didn't matter for what I was trying to do ( test the Blender posing) so I left it. However I have since started by telling Gemini to use the photo I provided and create a T-pose figure from that, and it has made a much better stab. I've finished the posing and its on the printer now. There is still a slight weirdness to one arm but again I think that's something I did.

Yes, I've had that duplicating effect. I uploaded two different photos, one a driver and the other a fireman but they both came out with identical faces

[philipy]

Ok, 90% there now. To recap,
Scanned a photo from a book,
Loaded into Gemini,
Gave this prompt: Please create a high definition, photo realistic image based on this photo. Keep the same clothes and face, show him standing with arms outstretched and legs slightly apart. Remove all background and show him on a plain white background with no shadows. The finished character should be 85mm tall

(Had to change the prompt a couple of times to try to get something like the correct hat, but I'll still have to do some surgery eventually.)

Saved the generated image and loaded into Meshy. Again I had to get Meshy to reimage 2 or 3 times and picked the best one. Then saved the stl and imported it into Blender.

Used the instructions from my tutorial to add a bone skeleton, doing the fingers is a 'mare but necessary for the pose I want.

When roughly what I wanted, I saved the drawing and it's stl and imported that into my Sketchup 3d drawing of the complete loco. Positioned the driver in the cab drawing and inspected to see what if anything needs adjusting. What I was trying to achieve was: Left hand on the regulator handle, Right hand on leaning on the cab side sheets and body leaning to the right looking forward. It's made more awkward because the trailing axle springs stick up through the cab floor and he stands straddling it with the brake standard right behind him!
Anyway, Having an idea of what adjustments were needed, I brought the Blender window up again, a tweaked his limbs, saved it, exported the stl and repeated the SU import to check. I needed to repeat this process several times to be satisfied.
This picture shows the final blender image on the left and the stl in blue inserted into the loco cab drawing, on the right. As I said, his hat will need some attention in due course, and the reason he is solid blue/black is because the mesh is so fine that you can't see the gaps!

Screenshot 2025-10-11 13.59.03.png (823.52 KiB) Viewed 67 times

Screenshot 2025-10-11 14.33.02.png (117.37 KiB) Viewed 67 times

Nearly forgot, I gave Gemini the instruction to make the image 85mm high, in the initial prompt. That has carried right through to the end, although I repeated it at the Meshy stl export to be sure. Blender's sizing/scaling is impossible but with an 85mm figure stl going in I got an 85mm stl out.
Now all I have to do is repeat for the fireman!

[Paul_in_Ricky]

Hate to point it out, but that model driver looks much shorter than the actual drivers on the F&B.

let's not mention the baseball cap

[philipy]

Yes, I know, but its hard enough to get him in the space as it is. I can control the height but getting width and depth and in proportion and still look human is a problem. So I prefer to think these two were a different pair of shorter, stocky Welshmen.
As I said the cap needs some surgery, but its easier to do that with a file after printing, than to use software of some flavour.