I wonder if anyone has any experience with this?

I decided to give Prompthunt.com a go. (It has a free option if anyone are interested in trying it out)

For those that don't know what it is, you can use it to make images by typing a description of what you would like.

So for example, I used this prompt for testing:

"Woman warrior with a sword and shield fighting a giant snake with long spikes on its back in a spooky forest"

It can create some really good-looking things, but it seems to struggle a lot with making specific stuff and in a lot of cases it seems to just ignore what you say or add stuff even if you specifically stated how many there should be. You also have the ability to add a reference image, which is what I tried with Poser.

  

So I mashed this together extremely fast, simply interested in the composition and posing of the characters. Through Prompthunt you can then apply different styles, technics etc. and it will create an image.

After a lot of tries (Im really new to it so there is a lot of trial and error), I got it to create this image, which was the first that was remotely good looking.

At first glance, it looks pretty good, but the shield is not there, and the tail of the snake seems a bit weird and doesn't follow the Poser reference very well. Which I think can be explained due to some settings, where you can tell it to track the pose. Which seems to work somewhat ok for the woman.

I then told it to use that image as a reference for future images and this is what I ended up with after changing a bit of stuff around.

Overall it did a fairly decent job, the characters don't really seem all that interested in each other and their faces are a bit off. But given the quick Poser draft image I used, I think it is pretty good for 10 minutes of work.

Have anyone played around with this? and know if there is a better way to get it to do what you want, like making sure that the shield is there when I specifically wrote it should and for instance, if I wanted them to look like they were fighting, is there a trick to do that? which it also seems to have ignored. 

These are just some examples I made trying to figure out some styles. I think it does a really good job. But I would really like to be able to control the scenes a lot better, instead of just relying on "random" stuff, but at least it is very useful for coming up with ideas for designs.

Even if you wanted absolute control, the AI will only work with what is available in its database until that database expands.  IMHO, trying to feed it Poser concepts is just piecing together random images from already established data on the internet, either legally or illegally.  I have only used AI to create textures or assemble conceptual art that I can recreate in Poser.  I can't justify creating art through AI when it 'borrows' content from someone else's art.

I don't really see a huge issue, as long as the AI doesn't directly copy/paste someone else's work. Is it any different than someone learning from looking at Leonardo's stuff or taking inspiration from someone else design in order to create something? It doesn't simply copy/paste different things and put them together randomly.

Besides that, the whole debate about whether it is one thing or another is way too late now, that train left the station a long time ago and is not coming back. It is impossible already now to know which images are AI generate, which are real or if they were created by an actual artist. 

 

How would anyone be able to do that, unless it is an actual copy? 

Trying to fight AI generate art is cool and all, but it is a lost cause and Ill rather be realistic and enjoy it for what it is and that it is simply how things are going to be from now on and then get as much out of it as possible, it doesn't even remotely ruin my passion for creating things, I actually think it makes it a lot more fun.

I make a lot of poses for V4 and M4, should someone be inspired by them and create them for another character, I don't really have an issue with that, I mean I didn't invent these poses I get the inspiration from photographers taking pictures of people.

Have found some information about it and it can be done, but is a bit technical. But this will explain how to do it, how to replace faces, poses, objects etc. https://stable-diffusion-art.com/controlnet/ 

Also, there shouldn't be any issue using Poser to do it as I can see, as there is already someone that has made a simple 3d posing tool that can be used. But Poser seems much better suited for this compared to that, so that is perfect :D 

Some AI-generated videos, really cool as well, I didn't make them. Be advised that the sound in the first video is pretty high so good idea to turn it down.

AI generated dancer

And another AI dancer in anime style

AI Anime dancer

Not sure :D

AI something

Making art is going to change big time :D

Firefly - Adobe

good.

AI-generated 'art' has its own dedicated forum, please keep this there and don't pollute the Poser forum with this..

Has anyone found one which improves/fixes hair?

Hair generally seems to be a 'giveaway' in my attempts at realism, with both Poser and Studio, especially with the free, basic hair that I use!

regards
prixat

All sounds too much like a 'Make Art' button to me, so I will stick with Poser and the only AI I will use is in some of the post work like anti noise.

 

 

I use Poser 13 on Windows 11 - For Scene set up I use a Geekcom A5 -  Ryzen 9 5900HX, with 64 gig ram and 3 TB  storage, mini PC with final rendering done on normal sized desktop using an AMD Ryzen Threadipper 1950X CPU, Corsair Hydro H100i CPU cooler, 3XS EVGA GTX 1080i SC with 11g Ram, 4 X 16gig Corsair DDR4 Ram and a Corsair RM 100 PSU .   The desktop is in a remote location with rendering done via Queue Manager which gives me a clearer desktop and quieter computer room.

Not to worry, all done! LOL

regards
prixat

hornet3d posted at 4:24 AM Tue, 27 June 2023 - #4468757

All sounds too much like a 'Make Art' button to me, so I will stick with Poser and the only AI I will use is in some of the post work like anti noise.

If you just want random stuff then I would agree, but if you are looking for something more unique or specific you are going to work for it :)

This is an example (Using Stable diffusion):

So say I wanted a man standing in a forest and just write this:

Prompt:

" Man with short black hair wearing a black shirt and denim pants is standing casually on an old dirt road surrounded by a dense forest"

Then this is what I get.

it's not extremely terrible but still, there is a lot wrong with him. And you have to fix that through writing. So I adjust the prompt:

"Man with short black hair wearing a black shirt and jeans is standing casually on an old dirt road surrounded by a dense forest, (Super realistic), dof:0.3, Natural sky, (Full body view)"

And then it gives me this:

That is pretty damn awful.

So having worked with it and changed things around and doing a lot of renders, this is what I ended up with and didn't wanted to work with it anymore:

"(Extremely details),(Extremely high quality), ((Extremely realistic hair)),(diffuse lighting:1.2),(Very realistic),(dof:0.2), (accurate reflections). (accurate eye reflections),(Natural colors), 54yo man tough with short hair and long beard wearing a(white stained shirt):0.5 and jeans standing casually, (background old extremely realistic dirt road surrounded by a dense super detailed forest),

(Randomize body view) , (Professional composition)"

It's still not the best, but a lot better. 

But you are working with two types of prompts, one positive which is the one I have posted above and then a negative one which tells it all the things it shouldn't do. Which end up looking like this:

"3d max, grotesque, desaturated, painting, cartoon, B&W, deformed, unrealistic face, unrealistic eyes, ((Deformed mouth)), deformed hands, blur, incorrect number of limbs,(disfigured, painting, cgi, illustration, photoshop, photoshopped, video game, 3d render, watermarks, bad art, Octane, multiple figures, duplicate, body out of frame, face out of frame, cloned face, weird colors, disfigured, bad anatomy, bad proportions, poorly drawn hands, extra fingers, malformed limbs, missing arms, mutated hands, extra limbs, missing legs, extra legs, poorly drawn feet, ugly, cross-eye, poorly drawn face, tattoos, text, writing:1), inaccurate teeth, inaccurate lips, inaccurate mouth, pixelated, low resolution, artificial lighting, oversaturated colors, flat textures, incorrect shadows, unrealistic reflections, unnatural proportions, distorted geometry, inconsistent scale, poor composition, repetitive patterns, lack of depth, excessive noise, fake-looking foliage, unconvincing weather effects, synthetic textures, unconvincing skin tones, rigid facial expressions, inaccurate object scaling, unconvincing environmental interactions, ((man not having a calm look on his face))"

And the way you write it, meaning both words used and the order matters, I would best describe it as talking to someone who is extremely stupid and can't figure out to do anything on their own. :)

There are places like Prompthunt that I used above which are slightly easier to get you a quick image, but it suffers from the same issue as Stable diffusion does which is that it is insanely difficult to control stuff. There are some tools to help you, but still its limited I would say.

It is extremely funny, but it is not just click to win unless you just want random stuff, which is fine if that is what one is after. 

But if you haven't tried it, I would recommend you give it a try it is not as easy as one would think :D 

With only the OP comment talking Poser I'm tempted to say this thread should be moved from the "Poser - OFFICIAL" to the "AI Generated Art" 😆

@3D-Mobster yes I think you are totally right when you say: "...but it suffers from the same issue as Stable diffusion does which is that it is insanely difficult to control stuff."

I can related to this specially when attempting some complex spatial effects, with specific shapes/forms/animations, I can't imagine at least for now, a system that can give us a vertex/pixel/point level control of creativity FAST. Except in the future maybe when our brains are connected directly to the machine!

See for example some 3d assets I have done, like 3D fire effects, the more complex it gets, the harder it is to explain to the machine, but if you do it the OLD way, like drawing/painting/modeling it is already faster in some cases.

---

EVargas.Art

evargas posted at 7:17 AM Sun, 2 July 2023 - #4469105

With only the OP comment talking Poser I'm tempted to say this thread should be moved from the "Poser - OFFICIAL" to the "AI Generated Art" 😆

@3D-Mobster yes I think you are totally right when you say: "...but it suffers from the same issue as Stable diffusion does which is that it is insanely difficult to control stuff."

I can related to this specially when attempting some complex spatial effects, with specific shapes/forms/animations, I can't imagine at least for now, a system that can give us a vertex/pixel/point level control of creativity FAST. Except in the future maybe when our brains are connected directly to the machine!

See for example some 3d assets I have done, like 3D fire effects, the more complex it gets, the harder it is to explain to the machine, but if you do it the OLD way, like drawing/painting/modeling it is already faster in some cases.

I could see the future of programs like Poser and Daz integrating AI into the workflow, where you wouldn't really use high res figures etc to make scenes, but more as design and composition tools. Because Poser offers an extremely easy way to pose characters compared to 3d applications and could you make an AI identify and replace characters and objects in your scene that would be really cool. So I think the "future" battle will be about who can make the program that is the most user-friendly. And honestly, I think Poser has an edge here in many cases. If something like this works, then it will completely blow all other current applications of the market, as the quality would be so much higher. And then it would be about who can make it work with videos, audio, the speed and having the best model to work with. I don't think either Daz/Poser/other posing program in their current form will have any future if they do not start to include AI. I think the question is more, who is the first to actually realize something like this?    

I'm still looking for something to fix Poser hair!

regards
prixat

prixat posted at 8:39 AM Sun, 2 July 2023 - #4469112

I'm still looking for something to fix Poser hair!

That is one of the issues as it is not, it often doesn't do what you want it to :D But this is a test I did using "Inpaint" in stable diffusion by masking the character face:

Original image:

And this is the result:

These were done using a prompt:

((((best quality)), ((masterpiece)), ((ultra realistic)),((extremely realistic hair))))

((((best quality)), ((masterpiece)), ((ultra realistic)),((extremely realistic brown hair))))

((((best quality)), ((masterpiece)), ((ultra realistic)),((extremely realistic windy brown hair))))

((((best quality)), ((masterpiece)), ((ultra realistic)),((extremely realistic neon hair))))

So it doesn't really follow the rules extremely well, since it also changes head direction and look of face compared to the reference image, which is why I think that if programs like Poser/Daz can integrate it and make it work it would be a game changer, because as it is now you have to accept it changing things you don't want. 

In your case, I think Firefly is probably better until the AI generation allows for more control. 

prixat posted at 8:39 AM Sun, 2 July 2023 - #4469112

I'm still looking for something to fix Poser hair!

I used a bit incorrect setting, you should use the option "Inpaint not masked" and then you mask the character.

This is the result:

Probably a good idea to read up on what the setting in "inpaint" does as im not really sure. Also probably spend a lot more time on it than in my example :D