Hi All.

I've been on Poser off and on for 30+ years. Every once in a while I get the urge to push for realism. I have the fabulous model V4-Sasha16  on board. However, I've never gotten where I want to go for realism.

But guess what, the face of a human being consists of a structure and a surface. I now use Poser for the structure, and AI for the skin and details.

Here is the abstract of the pipeline: I get V4-Sasha16 in the window, and spin up the character I want, male or female. I am big on asymmetry for this, including splitting morphs and giving left and right slightly different settings. I try to not use any expression morphs to set the structure. This is important for animation, to keep the foundation of the person one thing, her expression at a given split second another.

Once I've got something, I render out what's called a gray-clay image reference set, no background, no skin textures, no hair, simple lighting. It looks like this:

This is not a depth map. It is a preview render out of Firefly.

There are many AI platforms to take it the rest of the way. The one I favor is: ThinkDiffusion.com on which you just pay for time to tinker.They have a template for the popular LLM "Juggernaut/Wan2.1 + Control Net."

You load the Poser reference into one of the two ControlNet slots, and an image of any human face legal to use just to get the details, coloration, pores, eyelashes, etc. in the other. The final will not look like the person in slot 2.

The Poser ref drives Juggernaut/Wan to create an image locked onto it. You also construct a prompt for what you want to see.

Here is the result, driven by the above poser reference and a legal-to-use image of a woman's face to get the coloration and details. You can see the crooked nose and  other asymmetry that I dialed in.

A one-off Poser render to drive a portrait such as this is great. Rendering out an animation sequence from Poser is true power: the AI platforms are notoriously difficult to control precisely. Most of what you see is guided by a prompt full of wishes, but the platform has a mind of its own.

I hope this is not inappropriate to post in this forum. It simply delights me to know this function for Poser is so cool.

::::: Opera :::::

Ken1171_Design uses Poser to drive his AI artwork.  He has a couple of tools that might interest you.  In particular:  https://www.renderosity.com/marketplace/products/161157/openpose-for-poser-12

Molly

I just don't see the point. The AI output doesn't even have the same head shape. It's narrower, one ear is distinctly higher than the other (and not that way in the render), etc. If you're going to go for something that looks like it was produced by AI, why bother with the rendering step at all? It's not like you're achieving consistency that would allow you to, for example, tell a story with the character.

NOTE: my efforts are for animation, not stills. You can absolutely drive an animation with an image sequence out of 3D.

1) Once in the AI workflow, you can adjust the adherence. The issue is a) control of movement b) consistency of character.

2) And ..... realism --  are you rendering the human face with this amount of realism strictly in Poser across 2400 frames of a video?

3) Yes, people can tell it is AI. Just because you can't wholy defeat the Uncanny Valley (yet) doesn't mean you can't get story-telling power by getting close.

Note: 'one ear higher than the other' ... maybe the image I posted was frame 111, and she has tilted her head slightly. Also, you can see the bend in her nose in both above. The Poser output, uploaded into AI, respects the morphing.

Let me see if I understand this right...

- you tweak your Poser character (V4-Sasha16) to make for a more realistic and slightly assymetrical character (good idea, of course)

- Then you ask ThinkDiffusion to create a face, based on your tweaked texture-less head. (the result you've posted is the best one you got, I assume)

- How do you then go from your AI-generated picture to a functional Poser facial texture? Is there some clever app or script, that can 'snatch' a texture from a picture and turn it into a V4 compatible facial texture map?

Unless that can be achieved, I see the AI picture as just that, a picture...

...or is there something I have completely misunderstood? (And to be sure, that is entirely possible...) 

LapinDeFer, I start from Poser, but I don't return to produce my animation; I use AI for that.

I load into my AI workflow (called comfy.ai) ...

1) the gray-clay or depth map video as described above ....

2) I also load an image I generated in ThinkDiffusionJuggernaut, influenced by the poser output .... it has the underlying V4Sasha structure, with AI generated skin

The AI generation uses both references (plus the direction I supply as text in a prompt) to produce the finished video. The structure of the head or body is driven by the poser-created video, the skin and surface details are driven by the still image reference, and the prompt tweaks elements of the set, etc.

There are other ways to generate the depth-map or gray-clay animation, such as with Blender, but I happily use Poser for it, because of the excellence of the V4-Sasha model, and my long practice in Poser.

::::: Opera :::::

operaguy posted at 12:31 AM Sat, 5 September 2026 - #4510089

LapinDeFer, I start from Poser, but I don't return to produce my animation; I use AI for that.

I load into my AI workflow (called comfy.ai) ...

1) the gray-clay or depth map video as described above ....

2) I also load an image I generated in ThinkDiffusionJuggernaut, influenced by the poser output .... it has the underlying V4Sasha structure, with AI generated skin

The AI generation uses both references (plus the direction I supply as text in a prompt) to produce the finished video. The structure of the head or body is driven by the poser-created video, the skin and surface details are driven by the still image reference, and the prompt tweaks elements of the set, etc.

There are other ways to generate the depth-map or gray-clay animation, such as with Blender, but I happily use Poser for it, because of the excellence of the V4-Sasha model, and my long practice in Poser.

::::: Opera :::::

I am not really interested in AI videos...

Thank you!

I have to say that I'm not into AI pics/vids, given the insane amount of energy/water those data centers are destroying each time.