CAD figures for free by AI

[Paul_H]

If you’ve ever thought some 3D figures might be worth having, but given up trying to build CAD models yourself, you might like to try Meshy AI Community - Meshy

They currently have a seasonal promotion when you can get twice as many goes at generating a figure from a text prompt until 1/1/25. Free to register and a free account has enough credit to normally generate 10 figure sets (4 offers in each).

I’ve had a few goes at it and think it may be worth a play, if not an actual subscription.

Negatives;
Some figures are unusable with holes and hollow areas.
Some (free) figures are hopelessly low resolution.
The language model may not recognise some UK text properly eg bowler hat and flat cap were just ignored.
Some prompts just aren’t used properly eg a sitting figure prompt has generated a standing figure.
You’ll need to resize any figures generated to your preferred scale. That might help with resolution or not.
Worth noting it may be AI underneath, but as there’s no way of feeding back errors it’s not going to learn fast.

Positives;
You might get a unique usable figure for your railway.
Free

An example from this text prompt;
“Victorian workman wearing trousers, shirt waistcoat, bowler hat, short hair, Standing with legs slightly apart, right arm held slightly up to waist forward of body level, left arm beside body”
The photo shows what I ended up with.

 

[Paul_H]

Just to show it sometimes gets it hilariously wrong, here's the one it made with the last of this month's credit.
The prompt was;
"workman polishing locomotive with right arm out stretched, left arm held beside body. Wearing a waistcoat, flat cap, shoes"

 

[Phil O]

That should be in the Tate Modern!!!

 

[paratom]

Here is a 3D image of a railway man created from an old photo in Meshy which I intend to 3D print. Not bad and I suspect the software will improve over time.

 

[Osgood]

There seems to be what is either the front end or the back end of a 4 legged animal in that print - guess Al took Iron Horse too literally?

 

[Paul_H]

Not bad and I suspect the software will improve over time.

That looks pretty good. Better than the ones I've got from text prompts. Decent detail on the clothing too.

 

[david bigcheeseplant]

I created an STL file from the photo attached, its not exact but the programme has added feet that are missing from the photo. I have inserted the figure into my virtual model of Thame station. The STL file is created full size so needs scaling down for our required scale.
David

 

[simond]

are we at a stage where an AI system can create a 3D model of "common" shapes & forms to an accuracy sufficient for 3DP?

I'm thinking mostly of scenic things, like livestock (cows, goats, sheep, pigs) and other animals (dogs, cats rabbits, perching birds) and stuff that could be modelled directly like pillar boxes, bollards, motorbikes, luggage, parcels, fish boxes, wicker baskets, punnets, etc., but would be sufficiently detailed without spending hours creating them.

 

[Paul_H]

are we at a stage where an AI system can create a 3D model of "common" shapes & forms to an accuracy sufficient for 3DP?

I think it all depends on what scale you're working at and how high your standards are.

I've found the Meshy generations can look good on screen, but when printed lack a lot of detail (16mm/ft here).

 

[david bigcheeseplant]

Meshy created the following from the photo even including the dog, although has removed his beard!

 

[Paul_H]

The problem with these figures is that whilst they look good on screen, the actual STLs generated have very little detail. What look like folds, buttons and edges are only surface images on the screen render, it doesn't get resolved in the export.

 

[david bigcheeseplant]

Although not as much detail as a scan there is still plenty of detail on the printed models as attached.

David

 

[paratom]

Playing around with Meshy. Here is another image produced from a photo.