3D print people figures

[Phil.P]

Slightly concerned about the horses anatomy? - The legs don't look 'right' to me. But I am no anatomist.

The picture frame and seat cushion are very impressive.
Whether I could paint one, to match the quality of the print, would be another matter.

Now please stop!
I can't afford the time or money to go down this wormhole!

Phil.P

[Paul_in_Ricky]

I couldn’t persuade Meshy to create a slate stack that looked realistic so my effort was produced by getting Meshy to create an individual roof slate that I then flipped and mirrored a few times before randomly combing multiple copies into one STL.

The slate pile was an intriguing problem. It revealed some interesting failures with AI, both in Meshy, ChatGPT and Gemini. The systems didn’t seem to understand the geometry at all well despite being given what I’d thought were good photo prompts.
Not liking to be beaten I probably spent a good hour working through various prompts with different photos and edits in Photoshop and Chat GPT to get what I thought was a really good image, but Meshy just couldn’t generate anything from it. A final tweak to background colour made all the difference and it managed to finally deliver an acceptable STL, on the fourth retry, that could be edited in Blender.

Slate pile STL.jpg (179.9 KiB) Viewed 16477 times

The final pile STL prints nicely and is able to be rescaled to different sizes, chopped down to fit and, rather nicely, can be skewed to look like a stack of slate awaiting loading.

Slate piles.jpg (290.06 KiB) Viewed 16477 times

It’s even possible to cut an individual slate from the pile.

[ge_rik]

I like the way it's even represented the chopped bevelled edges.

Rik

[Trevor Thompson]

I have just caught up with the rest of you! Gemini and Meshy are amazing.

Screenshot 2026-01-21 at 12.04.20.png (536.55 KiB) Viewed 15711 times

I started last night looking for photos on the internet. By the following mid-day I had my first figure printing and a second ready to print.

It is a revelation to be able to create female figures appropriately clothed for the victorian period, that I can print out to go in my coaches - and of course so sit at the various stations.

I tried to use Blender and the latest version of Makehuman but I just couldn't get it to work properly on my iMac. Having spent days trying to sort it out I gave up and deleted it all. Within hours I am getting better results this way.

Trevor

[Trevor Thompson]

The result of my first attempt:

IMG_4945.jpg (2.89 MiB) Viewed 15688 times

and that is from a filament printer.

Trevor

[philipy]

The result of my first attempt:

IMG_4945.jpg

and that is from a filament printer.

Trevor

Wow, that is brilliant from a filament m/c. Well done and welcome to the funny farm!

[Durley]

She looks great Trevor. I’m still amazed what can now be achieved from this technology, allowing completely personalised and unique figures to be produced with so little effort.

[ge_rik]

I have just caught up with the rest of you! Gemini and Meshy are amazing.

Screenshot 2026-01-21 at 12.04.20.png

I started last night looking for photos on the internet. By the following mid-day I had my first figure printing and a second ready to print.

It is a revelation to be able to create female figures appropriately clothed for the victorian period, that I can print out to go in my coaches - and of course so sit at the various stations.

I tried to use Blender and the latest version of Makehuman but I just couldn't get it to work properly on my iMac. Having spent days trying to sort it out I gave up and deleted it all. Within hours I am getting better results this way.

Trevor

Hi Trevor
How was Gemini involved in the process?

Rik

[philipy]

Rik, I presume that Trevor is using something along the lines of what Durley outlined on P7 of this thread ( his post at 11-56 I think). That's what works for me anyway.

[Trevor Thompson]

Yes I am using gemini to isolate the figure from the rest of the photo and making the figure sit down.

Trevor

[Trevor Thompson]

She looks great Trevor. I’m still amazed what can now be achieved from this technology, allowing completely personalised and unique figures to be produced with so little effort.

I didn't expect the fringe on the shawl to print but it has most of the way around.

Trevor

[Paul_in_Ricky]

I tried to use Blender and the latest version of Makehuman but I just couldn't get it to work properly on my iMac. Having spent days trying to sort it out I gave up and deleted it all.

Makehuman has been sidelined by AI now. It wasn't very well supported and very limited.

I'd suggest you'll re-install Blender though. It's probably the easiest tool to re-pose figures to get them exactly as you need them. AI often gets the generalities of pose acceptably correct, but getting a hand precisely on a regulator or a head look in a different direction is almost impossible without editing the mesh (see philipy's instructions). I've also found that Meshy 6, at least, sometimes give extraneous detail or texture that is better being editing off the model.

[ge_rik]

Rik, I presume that Trevor is using something along the lines of what Durley outlined on P7 of this thread ( his post at 11-56 I think). That's what works for me anyway.

The first few posts on P3 explain the process as well. I might try giving it a go once I've finished painting the last batch of figures which Dean scanned for me.

Rik

[Trevor Thompson]

Rik, I presume that Trevor is using something along the lines of what Durley outlined on P7 of this thread ( his post at 11-56 I think). That's what works for me anyway.

The first few posts on P3 explain the process as well. I might try giving it a go once I've finished painting the last batch of figures which Dean scanned for me.

Rik

I have found that I need to edit some features of the models meshy produces. I find it easier to do that in Autocad fusion. I will only reinstall blender if I have to. I accept that exact posing it easier in Blender.

The advantage of cleaning up the photo in Gemini first is that I can get rid of extraneous furniture from the photo, as well as get it to sit the figure on a chair before removing the chair. Then I can get a standing figure and a seated figure from the same bit of photo.

Trevor

[philipy]

I've also found issues with bulky clothing, particularly with females in period clothing, if I just tell Gemini to show a figure sitting on a seat. Also it depends on Gemini's definition of a seat and the dimensions it uses, which are unknown, but I now think I may have found an answer to both of those points.

I need figures to populate my F & B coaches in due course and I have produced some individually by simply asking Gemini and Meshy as discussed previously, but they don't look quite right because they don't fit the seat accurately.

Trevor's brainwave of overlapping figures gave me an idea, so I took my Sketchup drawing of the bench seat from the 3rd Class coach and took a screenshot of it on a plain background. I then put that picture into Gemini and gave it this prompt:


This is a drawing of a railway coach seat, Please create a high definition photo-realistic image of two Welsh quarrymen of around 1900, sitting side by side on this seat which is 4ft 6 inches long. One figure should be slightly taller and be wearing a bowler hat. The second figure should be wearing a battered trilby hat. Both should be wearing hob-nailed ankle boots and their clothing should generally be work worn. One of the figures should be sitting with crossed legs, and both should be reading newspapers.

This is what it gave me:




I then told it to remove the seat and all background detail, downloaded that and uploaded to Meshy as usual. Meshy gave me this (seen from below, the flat undersides show they are actually sitting on the seat)

2 men on seat meshy.jpg (94.43 KiB) Viewed 15175 times

The Gemini prompt needs a little tweaking but it isn't far off what I intended, and Meshy seems to have got the right idea as well.

The thing that still eludes me is how to scale the downloaded Meshy stl in the slicer to accurately fit the seat when printed. The best I can think of is to put the seat stl into the slicer, together with the figures stl next to it, and juggle their scale until it appears to be right, then delete the seat. Chitubox won't allow stl's to overlap on the bed so I can't actually put them on the seat.
Has anyone got any other ideas?

I can't print them atm because I don't have a Meshy6 sub currently, that will have to wait until I have enough figures ready to justify the sub for a month!

Edited to add another bright idea - I told Gemini to remove the background detail but retain the seat. Meshy has followed suit and this is the result, so if it prints OK in due course, it should solve the figure fitting issue and also save on print time.

2 men on seat.jpg (164.29 KiB) Viewed 15171 times

[Durley]

Great result Philip. I agree printing the figures and seat all as one is an ideal approach if the seat geometry and support approach allow it. The figures can then be made to settle down to really look like they are sitting in the seat, rather than perched on top.

The slicers I mostly use (Halot box and Lychee) have a measure function so you can get a measurement between any 2 points on a model, I guess Chitubox does too? I have used this to measure the back of the knee to the base of the foot and then scaled the figures to get this measurement to match the seat height.

Incidentally, Lychee allows you to position multiple STLs anywhere in space and then combine them, that gets around the issue where each individual STL is forced down to the build plate. Blender has a similar function, but I find Lychee more intuitive, however I think the combine function is only available with a Lychee paid subscription.

I note Meshy has now moved to ‘Meshy 6’ I.e. no longer at preview. From a quick play about, it appears to be better at solid non-organic objects than the preview version, giving more pronounced structure to the objects. I was getting some promising results creating a pot belly stove to go in a brakevan for instance (as long as you don’t try to read the makers name…)

IMG_5749.jpeg (29.55 KiB) Viewed 15148 times

IMG_5752.jpeg (129.18 KiB) Viewed 15148 times

[philipy]

Wow, that stove is amazing!

The pictures I posted above are all with the full v6. I did try using v4 which gives free downloads but the quality is Meh!!

I've got Lychee but never really got on with it, not sure why. Plus I'm not sure that it can produce .m7 files which my Photon M7 requires. I haven't looked recently so they may have included it by now.
Yes Chitubox does have a measure function and I've tried using it to scale sitting figures to sit on a seat, but I'm never quite sure where the measure points physically are on the model.

[philipy]

It appears that I was only half right... Chitubox does have measuring tools, but only with the paid-for version. Unfortunately, the same thing applies to Lychee as well, and to me that little bit of convenience isn't worth 10 quid a month!

[philipy]

Don't know if this might be of general assistance?:

I've been trying on and off for ages to create passengers for my F&B coaches and try as I may, Gemini insists on giving the quarry men long lace up boots which simply don't look right to me. It has created these three on the seat and two have the long boots and one has 'normal' shorter ankle boots, even though my instruction was for all to have the shorter ones. This has been going on for ages irrespective of how many figures I try to create in the group or what wording I uses..

original.jpg (229.86 KiB) Viewed 14224 times

So this morning I asked Gemini what Prompt it needs to be given to create the boots on the RH figure and this is the reply:

To get the AI to consistently replicate the specific look of the right-hand figure—where the trousers break naturally over the laces—you need to use descriptive "anchor" phrases.

The model currently keeps defaulting to high-shafted "riding" or "trench" boots for the other two men. To fix this, you should emphasize that the trousers are straight-leg and un-tucked, and the boots are short ankle boots.
The "Pro-Level" Prompt Formula:

If you want to recreate that specific look in a new image, try using this wording:

"Three 1900s Welsh quarrymen sitting on a wooden bench. All three wear heavyweight, straight-leg charcoal wool trousers that drape loosely over their footwear, ending at the ankle. They are all wearing rugged, low-profile black leather ankle boots (work bluchers) with visible heavy metal eyelets and thick, mud-dusted soles. Crucially, no boots should be knee-high; all trousers must remain un-tucked and extend down to the tops of the feet."

revised.jpg (231.8 KiB) Viewed 14224 times

Why this works:

"Straight-leg" & "Un-tucked": This explicitly tells the AI not to tuck the pants into the boots, which is why it was generating those tall, military-style boots for the first two men.

"Low-profile ankle boots" / "Work bluchers": This specifies the height of the boot. "Bluchers" or "Derby boots" are the technical terms for that specific 1900s work-boot style.

"Drape loosely over footwear": This creates that realistic "break" in the fabric you see on the right-hand figure.

[/i]

It's interesting that Gemini can tell you what it needs to be told, and worth remembering for other input. I've wasted ages trying to do things simply because I didn't use the right words. Obviously in AI, syntax is everything.! The other point is that personally, I've never heard of "Bluchers" or "Derby Boots" as types of footwear. I suspect they may be Americanism's, but worth bearing in mind in general that when trying to write a prompt that doesn't give the intended result, it may be wanting to speak American not English!

[ge_rik]

Interesting. I wonder how long it will be before we all start talking to each other in AI-speak .....

Rik