A Face Thing, for Poser.  That is, set some dots, with a bit of help from dlib, if you want it, and then press a button.  And you get a morph!

It's not quite at the "Eureka!" level of success yet, but it's beginning to give encouraging results.  Some may be familiar with a certain Face Thing that has you set a bunch of dots and then always, always spits out the elephant man.  This is working better than that, at least.

I'm not very good at the internet back-and-forth, lately.  But sometimes I get excited enough about something that I want to share a bit.

Anyway, if I can get it into shape, it will be free, for anyone willing to load Python 3, Panda3D, dlib, and dlib's required Pillow and Numpy.  Is dlib capitalized?  I don't think I've seen it capitalized.  I'll leave it lowercase.

I guess that's all.  I'll show you some screengrabs, then stand here, twitching nervously.  Well, the screengrabs are all out of order, but... meh.

Sweet.

Cool...if I understand correctly similar to a face transfer for DS. Very cool.

Hey, Cage!  This looks pretty cool.  It's been some long time since anyone has gotten something like this going for Poser.  A long time ago I had a program like this that sort of worked and subsequent upgrades pretty much made it useless (Elephant man.... Heh.  I like that XD ).

Good luck, I am watching with great interest!

Teh Mollz

this looks really useful, nice to see your work again.  hope i can figure out how to install it when you get it ready.   btw i really miss being able to use your old scripts in P13 (esp chain maker (the one that places markers & makes a chain) & mr. skinner);  too lazy to reinstall P11 on my new rig though it may come to that.

jancory posted at 7:00 AM Fri, 30 January 2026 - #4503548

this looks really useful, nice to see your work again.  hope i can figure out how to install it when you get it ready.   btw i really miss being able to use your old scripts in P13 (esp chain maker (the one that places markers & makes a chain) & mr. skinner);  too lazy to reinstall P11 on my new rig though it may come to that.

There are days when everything goes as anticipated, when programming a thing like this, and then, for me, days when I haz the dum and everything I try fails maddeningly and incomprehensibly.  I've had those two types of day back to back.  Grumble.  The core of the program works, it could be used to produce morphs now, but it has no undo function and can't save and load projects-in-progress.  It will be able to save the marker setup for a picture set or a Poser head, so these can be easily reloaded for later projects.  But that isn't set up yet.  Right now, I'm trying to de-skew front pictures with the head turned slightly, using an approach similar to the parallax effect seen in some cartoons and video games.  But, the dum haz me, alas.

i had P13 on a very outdated win7 machine til the machine pooped out.  mostly ran fine.  a few things didn't work for me that worked for others so it may depend on how updated your win7 is?  had the machine stayed alive i'd still be using it.  

All of that hinges on the idea of Poser 13 being willing to run on one of the klinky, old Windows 7 machines I use for Poser.  If that doesn't work, my plan may be sunk, at least for a time.

Poser 13 runs on Windows 7 for me - although, like you, I only just got it so I can't claim that every last feature works.

For what it's worth, I have to go back to Poser 7 to use Mr Looper et al. It throws errors on Poser 11, but that's been updated several times since I bought it from Smith Micro, and reinstalled at least once. It still isn't feeling very well, and may have to undergo major remedial work along the lines of "wait until the thunderstorm is at its peak, and then throw the switch, Igor!"

P13 was pretty stable for me on win7.  more solid than win11 has been so far for me---i can make P13 go poof & disappear without a trace  doing some of the most mundane things on my current machine.  i'd go back but i've already stripped the drives out of the oldie plus this one has a graphics card so renders are fun again.

sorry to sidetrack your thread, cage.

Sidetracking is okay.  I do that, myself.  One reason I don't use later versions of Poser is that phone-home dealie.  My W7 devices typically don't touch the internet, and haven't since Microsoft won my distrust while they were tricking people into upgrading to P8.  Properly borked my mother's laptop, and there's some sidetracking, eh.  I only use Linux devices for the internet, but the only Linux device I have right now that isn't a Chromebook (and thus not Poser-friendly) is itself pretty old and klinky.  Nonetheless, I'll see what I can do about script updates.

understood & totally agree; imho the phone-home thing was a big reason we lost so many users when they put it in, what, P7?  before that poser was growing & thriving.  i understand the need to limit bootlegging but i think it stifled growth.  we really need an offline option.

anyhoo, thanks in advance for whatever you can puzzle out.

It wasn't until Poser Pro 2014 Game Dev that they added the phone home. Poser Pro 2014 and Poser 10 didn't have it. 

thanks, faulty memory here.

Smith Micro was the one to add the phone home. When they killed the servers, I had to roll back to pre-game dev to be able to continue using it. Poser 7 was the last version before SM bought it. 

That's why I still use PPro14, yup.  I always figured Steve... Cooper?  quit in protest over the phone home thing, but who knows.

You see, about the off-topic drift?  Oh man.

The parallax thing didn't work, but so far it looks like I've achieved the desired results using some rotation matrix tricks.  I'm not sure it's fully de-skewing a face, but it is putting the nose back in the middle, which makes morph cleanup using symmetry tools much more approachable.

Something I am learning about tools like this one is that the selection of source images is very important.  One may think one's front-view image is pretty straight-on and doesn't have the face rotated at all toward a 3/4 view, but that may be happening more than one realizes.  Every human face has asymmetries, but they can end up distorted and exaggerated by a tool like this, so the inputs are important.  And lining up the profile view image with the front is tricky.  The test I show at the top o the thread has a profile image that is misaligned with the front view.  I suppose I should think about ways to compensate for these things, even after the images have been prepped.  Hmm.

No one knows what you're talking about yet.  Stop babbling.

It's coming along nicely, but still needs undo and proper save before being ready for testing.

Note my typo, way up there: Microsoft was tricking people into upgrading to W8, Windows 8.  Not P8.  That was Smith Micro's Poser release that was dominated by GUI alterations, bringing in the Internet Explorer and Flash dependencies.

Script stuff is beyond me but the concept I understand, like front and side views matching. I had attempted to use the old face room at times. I could never get good results

The matching is important.  I think this process needs to allow the user to go back and make adjustments.  The reference pictures are imported, and the user can rotate, scale, and position them as needed, with reference points provided to try to line them up.  That works fairly well, but the nose is often still too high or too low.  A profile image can have the mouth and eyes line up decently with reference points, but the nose can be off, perhaps in part because the mouth is all muscles and thus an unreliable landmark.

But once the basic marker points are auto-placed from the front view, they are projected back on Z and (roughly) auto-positioned on the 3D head.  Once the Z-locations for those are finalized, those can be copied over to the picture side.  At that point, the profile view for the pictures has the Z-positions from the head and the Y-positions from its own

front view.  Which offers a better way to align the profile facial elements with the front.  The user needs the opportunity, at that point, to step back and realign the profile pic so it better fits the Y-locations it inherits from the front view.  Then the user can go on to set the Z-positions for the profile pic.

I think that probably sounds more complicated than it is.  A lot of this is more-or-less automated, just requiring tweaks and corrections.  The process I outline here should help better-align the two reference images, at least.  

Wow, nice!

Here we go.  Some pictures.

The reference images have been imported and positioned.  Here, the dots have been created and auto-positioned by dlib.  I've made some adjustments around Antonia's upper lips, because it always places the yellow dots too low, for some reason.  Possibly due to the lighting setup I'm using.

From the side, we can see that some dots projected back along Z landed in places that need adjustment.  The picture dots can be seen to have Y positions defined, but no Z.

Adjusting the Antonia dots, then copying their Z positions over to the picture side, we can now see whether the picture's profile lined up decently with the front.  It isn't too bad, here.  The blue nose dots are where they should be and the purple eye dots aren't badly placed.

Adjusted dots for the profile picture.

After that, a morph could be created.  I spent twenty minutes or so going back and forth between the Face Recognition tab and the Comparison tab, making tweaks to the picture's dot positions.  Is the result good?  I dunno.  It seems to be working pretty well, but I can't always tell whether a face looks like another face.

This author, as seen 21 years ago....

I've been adding dots, raising the dlib default of 68 up to an optional 112, and I think I still need to add some for ears.  The final release will allow the user to decide which of my extra, added marker dots (if any) to use.

I've been working on file i/o and texture projection.  The main comments I've been getting about any level of resemblance in the results tend to focus on nostril shaping, so I still need to refine the marker setup in the nostril area and, I think, the chin.  Oh, and the ears!  So the 112 will end up around 124, in the end.

If anyone is out there, you may be asking, "But Cage, from whom are you receiving feedback, given the resounding silence, here on the forums?"  To which I reply, "Ha-ha, yes."  With these two I am now out of pre-existing front and profile image sets with which to test.  I'm going to have to gather some more, I guess....

what would be useful for this is seeing rendered textured images. without those it is a bit hard to opine as an end user in my opinion.

i wouldn't be able to do anything with it anyway, it has too much fiddling with things (python, dlib etc.)  I have no idea about... so feel free to ignore me.

Fair enough.  That sounds a bit like the "soft criticism" I've been encountering a bit, here IRL.  "Well, it kind of looks like, but.... merrm...." and they trail off into mumbling and can't be made to commit to solid, coherent feedback.  Dammit, I need my old emoticons, to show when I'm laughing at an idea and not ranting.

I'm accustomed to fighting with the results of that *other* FaceThing, using magnets and scripts to make the tiny bit of resemblance it could provide into something usable.  If nothing else, the results right now can make that simpler for me.  I may also have lower expectations than many of the remaining Poser users.  I'm not hung up on photorealism.  Which is to say, I'm not interested in it and I couldn't do it if I actually wanted to.  So I can't give you those kinds of renders, alas.

Anyway, I'll keep trying to refine the results, and perhaps I'll have some Poser views of exports, at some point.

I actually came back here to note that these basic procedures -- using marker point correlations to bring two things toward the same shape -- could be useful in other contexts.  What are joint centers and endpoints, but useful marker points that dissimilar figures have in common?  I think the basics here could be made into a tool to, at the very least, significantly simplify garment conversion using the whaddayacall Fitting Room.

For what it's worth, it looks to me like you're getting some pretty good results for shape of the head and facial features.  The production of a texture?  That seems to me to have a bit further to go than getting shapes right.  The idea of using technology like this for replacing the fitting room is very interesting, to say the least. =)

Teh Mollz

Oh, no.... the texture projection could not really be anything more than a visualization tool, perhaps solely within this face-shaping script itself.  The process is working with 600 x 600 pixel reference images and producing a 256 x 256 texture on the heads.  No, no.  This is not a texture making tool!  Yikes.

I don't know about "replacing" the fitting room.  "Augmenting", perhaps?  As seen here, a, what?  5500?  polygon head can be shaped into a decent approximation of a resemblance using less than 150 marker points.  After that, mesh cleanup may be needed, and the finer details of a portrait will need to be approached with more precise tools.  The same would be true, if this sort of approach were applied to garments.  It could provide a nice starting point, maybe help you jump into the process in the middle, instead of at the agonizing start.  Like that.

The answer is 6487 vertices.

i've been watching because i know from experience that sometimes Cage goes off on some very interesting tangents & ends up with some really useful tools.  that sometimes i can figure out how to use.  forward into the future!!!

i've been watching because i know from experience that sometimes Cage goes off on some very interesting tangents & ends up with some really useful tools.  that sometimes i can figure out how to use.  forward into the future!!!

Fair enough.  And I am a paranoid freak, so there's that.  That's what you're SAYING, ISN'T IT??!?!  CAGE IS A PARANOID FREAK??  I knew it, everyone is calling me a freak.  Why do all the people hate me, I must smell funny.  I'm well-practiced at that sort of thing.  One might say it suits my WEIRD, LOPSIDED FACE THAT NOW EVERYONE CAN SEE ohman never post things like that to the internets, now all the ai bots has your face from two decades ago and they'll use it nefariously!

... And this is how one makes people stop listening.  My apologies.  I don't mind lurkers.  Just, if you have any potentially constructive input, like "All of those still look like Antonia to me!", or conversely "These are all distorted to the point of caricature!", it could be useful.  "What do the meshes look like WITHOUT hiding them behind textures?"  "How much do the eyesockets drift from their original locations, with these morphs?"  "I hate Antonia, with the nose.  Show us Lafleufleu!"  "Those source images have their noses way off to the left, but the morphs are pretty symmetrical.  How does that work?"  I dunno.

Anyway, I'm just glad to know I didn't kill the forum.  I posted four images, one night... and then the forum became inaccessible to me for two days.  And then-then, now one posted at all to the Poser Forum for days afterward!  I figured, "You've finally done it, Cage, mboy!  It took you 25 years, but you finally killed the Poser Forum!  That's quite a step up from being an accomplished killer of mere discussion threads, to kill an entire community!"  Being a paranoid freak, and all.

Also, being one of the tiresome folks who always try to use "fewer" when appropriate, I was saddened to realize that last night I had used "less" when I meant "fewer".  Dang.

I just finished rewatching Freeman's Mind, and 9+ hours is long for any machima, so I read the first paragraph in his voice. 

Maybe a more productive comment will be that since you've been using only Antonia, can it do other figures? 

It is designed to be used with any figure at all, but there are many things yet to finalize before I can start varying the heads along with the pics.  If I've done it right, one should be able to open a full Poser figure geometry and receive a morphed head as the result.  Proof of concept tests for those features have worked with simple test geometries, but I haven't reached the point of testing with a full figure geometry.  

Some more screengrabs of development progress....

These show a random mugshot gathered from a google search, and an extreme stress test for the process.  The markers interfere with one another, beyond a certain point, limiting one another's influence on nearby vertices.  Usually I run with three markers able to influence any given vertex, but Gollum was so extreme (dlib didn't even recognize it as a human face) that I had to reduce the influence to just one marker, then clean up a very rough transfer with three passes of Restore Detail (seen here as 'Clean Up').

Finally, this is how the meshes look in Wings3D.  They're pretty clean, but you can see how the extreme Gollum morph futzed up the fragile relationship between Antonia_1.2's face and brow layers.  I have another, rather involved, process that I'm going to integrate for that sort of correction.

Also note that the areas outside of the marker ranges (such as the scalp and underjaw) still need some kind of handling.  The process is protecting the weldline vertices, but still needs various other refinements.  I'm thinking of adding an Adjustments" tab just for the cleanup tools.

Wow!  Thanks for sharing your progress. o.o

Teh Mollz

A quick screengrab that shows the range of results I've managed from Antonia, before moving on to extend the process to work with other geometries.

And here we start with Pauline.  I wanted to try LaFemme (?) first, but I haven't installed any of the Bondware versions yet.  I did have my old Smith Micro Poser 11 Beta installed, so I grabbed what I assume was the new figure for that version, for use as my first test of other head geometries.

An interesting irony lies in the fact that it really, really looks like Pauline's face was based on a generalization of a young Romy Schneider, since I have been using that actress as one of my basic testing faces.  My test became an effort to make Pauline Even More Romy-er.

I imported a saved comparison for Antonia, and a manual conversion for the new head was fairly simple, using the tools I already have in place.  That's encouraging, as my plan has been to try to automate the basic conversion of pic ini saves to work with new heads.

There are also some complications.  I'll continue....

the antonia morphs show great promise, looking forward to seeing how they do in renders.

That was Pauline, out-of-the-box, up there.  A lot of facial similarities, really.  I would be surprised if that wasn't intentional.

Being apparently based on that actress, Pauline has what we might call a Significant Forehead.  It's significant like Antonia's nose is significant, a feature that is arguably more pronounced than one should be for the shape of a Poser head.  One would think more generalized and bland would be better, but we keep getting base shape faces loaded with personality.  Anyway, I need to compensate for that sort of thing, with my process.  And one thing I note is that I don't have a good way to diminish a brow and forehead like Pauline's.  The process as it stands can metabolize a prominent nose or weak chin, or presumably any outsized or overly-prominent facial features from the middle of the face or the jawline, all those areas you can see covered with dots, in the above image.  But not the forehead.  So I need to work on that, and try to anticipate other potential geometry idiosyncrasies as I continue development.

The Pauline test went well.  Processing was quick, even on the clinky GalliumOS CELES Chromebook on which I'm developing.  It may be worth noting, though, that both Antonia's and Pauline's heads lack internal mouth parts.  I anticipate having to screen those out, once I test a V4 or something, and work out a way of coping withe the teeth.

These show some results from that initial test.  You can see that significant brow, and it makes me wonder not only how to reduce it with Pauline, but how to intensify that effect with the R.S. face for Antonia or other figures.  Working only with front and side images, something like that brow can be overlooked, since it seems mainly to be evident in 3/4 view.

I have other observations, but I'm not sure whether anyone is even interested, so I'm-a stop typing now.

jancory posted at 7:21 AM Thu, 5 March 2026 - #4504395

the antonia morphs show great promise, looking forward to seeing how they do in renders.

As for importing a morph into Poser and just showing how that works... yeah.  I haven't done much of that yet.  Just enough to verify that I'm exporting working geometries with correct vertex-ordering.  My excuse, if you want one, is that my desktop computer workstation area is SO BLANGED COLD in this house, during the wintertime, that I've been following my cold weather workflow.

The script's process is built around alignment of the eyes, and I've spent the last three weeks working on eye-anchoring techniques, along with symmetry tools.  So I know the morphs work and I know the meshes are clean and the shapes are not bad, and I know that the eyes should line up.  All of which suggests that we have the foundations for some useful morphs.  Poser is always full of surprises and aggravations, though, so I guess I'll see....

poser can be a harsh mistress.

impressed you're developing on a chromebook---all mine can do is access the internet.