Daz 3D is part of 
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.Daz 3D is part of
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Like this one, hands still need improvement !!!
No kidding! Very, very impressive.
I wonder if you draw stick figures and scene props what would it turn out like
but these are looking carraraliciuos yet hard to get my head around as to how.... I am flummoxed!
Very good, reminds me of the old great SciFi pulp covers.
Very cool, Diomede! Funny... I just watched Kong the other day! :)
Thanks for the feedback. After the Thanksgiving (US calendar) holidays, I will gather and organize some basic 'how to' info for a Carrara plus AI workflow, from installation videos to examples of a couple fucntions. My 'no one asked me' thread will probably be the destination. Meanwhile, I have another Brash example, which I will post in the Brash thread. The workflow for this specific image is posted in my Brash thread; I will put a more generic example in the 'No one asked me' thread.
https://www.daz3d.com/forums/discussion/comment/8454891/#Comment_8454891
Carrara Multipasses and Stable Diffusion
remember Carrara depth passes need inverting
it might make a difference to background and foreground masking
Correct! And even better, there is an in-painting tab for image to image with a check box to invert the depth pass. And there are control sliders to specify the amount of the effect for the foreground and the background. So you can load Carrara's depth passes instead of the one calculated by Controlnet (the Carrara one will be more precise, which might be good or bad depending on the effect going for). Or, you can 'fool' the Controlnet depth pass by having Carrara render out an object index pass, then use an image editor to make masks for various objects in your scene. Then you can load your custom masks in Controlnet and it will think they are in the background (or foreground) to include or exclude from further processing.
I will be putting together a screenshot demonstration of loading Carrara-generated masks in Controlnet after I get done with American Thanksgiving holiday traveling.
On drawing simple shapes.
Regarding hands.
There are already threads for beates over AI, this is not intended as such but as a showcase - if you disapprove of AI simply scroll on by.
One more Brash Lonergan example.
- I assembled a Carrara scene using my own custom models for Brash, his props (suitcase, etc), and the spaceship. I used primitives to block out the runway and ground plane. Emphasis. Everything in this scene was modeled, uvmapped, and rigged by me, or is a primitive.
- I rendered a main render, a coverage pass, an object index, and a depth pass. In this particular case, I only used the Carrara main render, but that depends on the project.
- text to image, with Carrara render loaded in Controlnet for 'canny' and for 'depth.' Note, in this case I am using the Carrara main render, not the depth pass. I let Controlnet derive depth based on my image. Sometimes, the Controlnet calculation is too fuzzy, sometimes the Carrara multipass has too little variation. It just depends. Here are my positive prompts to go with my loaded image. Positive. high quality comic panel full figure view
of a man blonde wearing a militaristic uniform (blue pants, red jacket) walking toward camera carrying a suitcase in right hand and a bag in left hand. Background science fiction jet on runway. In 1960s comic book style. Negative prompts. nsfw, low quality, signature, logo, watermark, speech bubble, child, children, immature, bad hands, mangled hands, extra fingers, missing fingers. -
Here is one of the 4 initial results
P4 Dork and P4 Posette are Management
Here is another example of using the old Poser 4 figures. I also used an object from Carrara's native content browser, the vending machine from the city folder.
Rendered out a simple scene using primitives as the floor and back wall, along with a multipass render for the depth pass.
In text 2 Image, the prompts were
Positive - high quality portrait of 2 people walking toward viewer conversing, in front of a vending machine
Negative - low quality, bad hands, bad fingers, extra limbs, extra fingers, missing limbs, missing fingers, nsfw
I used 3 Controlnet functions
1) Open pose to detect the basic poses of the two figures
2) Canny to detect the outlines of the two figures
3) Depth - I uploaded the Carrara multipass instead of having Controlnet auto detect
Here is the simple Carrara scene setup.
Here is the base Carrara render
Here is the AI output on first pass. Notice how improved the hands are compared to my first example of P4 Dork coming out of the graveyard. Still not perfect hands, but improved.
Here is a video that discusses some of the issues of multiple characters in a scene. Using a 3D program like Carrara and its depth multipass, or Daz Studio and its depth canvas, can help generate the Controlnet pose and Controlnet depth. My vending machine example above uses only Controlnet and Carrara. However, it is difficult to use only Controlnet and give different prompts for the two figures, such as if you wanted one to have blonde hair and the other to be bald. The hair prompts would interact on both figures unless more is done. There are extensions for that. This video discusses two other extensions that are specifically for images containing more than one figure, or more than one key landscape element, or more than one.... The issue is directing the AI to apply different prompts to different regions of the image, or to different elements of the image. The use of capitalized AND doesn't always yield good results.
Beautiful work Ted. Love the 1950's feeling! As ub said before he was 1984'd? I don't use Carrara for image making anymore, specifically because of ai. Not saying ai is a bad thing. It's an amazing tool for the tool box. After all, it makes us all into artists. As long as we can spell :) In the long term It bades ill for daz itself though - I think. Not that that's any of my business. At the moment I am doing things in the real world with carrara that ai can't steal - yet - ha ha. Not that there is a real world :) Keep up the beaut work.!
Appreciate your insights, Andrew. I am more interested in combining AI with stuff I create than doing pure text, but time will tell on that sort of thing. I intend to label AI in the title of any thread I start in which AI tools are possibly part of the discussion.
Hello Headwax - your renders will be missed !!!
ah thanks, the vacuum will be filled pretty quickly I imagine :) for example your beautiful work in the NPR thread Currently I am using carrara for tweaking surfboard fins...
Thanks, hope you find time to drop in and say "Gidday" !!!
How brittle are the Surfboard fins, can they take a bit of punishment ?
You always say nice things, which is why everyone likes you. :)
But no, the vacuum is not going to be filled. I cannot adequately express my admiration for your efforts. While many folks here have influenced me in positive ways, your art has been the biggest motivating example for me the past 7 years. But there are cycles in the human experience. We can all reach a point where it becomes obvious that it is time to move on. Best wishes!
I am trying to be optimistic
after all digital artists were shunned by Fine Artists for years as were photographers before them
then Poser users and subsequently DAZ content users
Diomede models his stuff in Carrara, AI is just another tool in the creative process
we are all producers in one way or another
Andrew wants to go back to the old school after using Carrara for years and I respect that
interestingly many of the AI groups I follow on Facebook are full of painters, crafters etc who are using AI for inspiration to create real world art and crafts too
I am the worry here in Carrara land as I have been using DAZ studio almost every day
as has Dartanbeck
we are the traitors
You are no traitor, Wendy. Nor is Dart. People use GIMP and Affinity and Photoshop. People use Filter Forge. People use UVUnwrap and UVMapper. People use Hexagon and Blender and ZBrush. People convert motion files and terrain elevation files. Of course people use Daz Studio and Poser. They are useful.
Andrew, cheers. I agree with UB that no one will fill the vacuum. Your contributions are beyond amazing, and your patience and understanding for untrained people like me is truly irreplaceable, and worthy of my highest thanks.
I respect that Headwax and UB believe they see the signs of a coming disaster from AI. It is frustrating to see something so clearly, and yet reasonable people fail to see it. I have had that on big topics twice in my life. Once was about 15 years ago and seems like Earth was an entirely different planet. That circumstance played itself out - there was indeed a disaster, but time has passed and we are all on the other side of it now. Even though I turned out to be right, I did not have a solution, and I cannot say the people who did not believe me were being unreasonable. The second time is now on a topic other than AI. This second potential disaster has not played itself out yet. I hope I am wrong. I have been wrong many times, of course. But these two topics are different from the 'many times.' Occasionally, there is something special - seeing something with enhanced clarity. Andrew and UB are among those who feel that enhanced clarity with AI.
For the moment, I am unsure about AI in its various forms. As Andrew pointed out, we already use it everyday in many ways. As a former fact-checker, no one has to tell me how awful the current versions of AI fact checking are. Yet I try to ask, compared to what? The other thing about being a former fact-checker is that I got to see first hand how often the experts are wrong (including me, ouch). Is AI wrong more often? More egregiously? With more deference? With greater consequences? These are all good questions, but I don't know the answer is that AI is wrong more often than the typical public library reference desk (um, that would be me in a way).
These are mere musings. I intend to experiment some more with combining 3D with AI in a variety of ways. I'll try to keep it out of the Carrara Forum unless there is some obvious Carrara hook - like these Poser 4 Dork experiments.
On the potential for AI to cause problems, on having that enhanced clarity about oncoming disaster, have a song. Bad Moon Rising? No. Something from the traitors.
Dork is Back - As an Old Man then as a Young Woman
More use of carrara to block out simple shapes as first step in an AI workflow - stable diffusion
The setup = Poser 4 Dork Casual posed in simple running in foreground, Poser 4 Lo man running in background and some primitives as a footpath
I rendered out multipass for depth pass in addition to simple base render
I used controlnet 'Depth' to process the Carrara depth pass to a SD depth image
I used controlnet 'Canny' to generate outlines for SD
I used Latent Couple with AND (all caps) to assign prompts to the background, left side, and right side of image.
Repeat, for prompt, AND is in caps - separate the (1st part = all image so state background), AND (2nd part = left half image), AND (3rd part = right half image)
You can actually be more detailed in dividing up the image how you want; but the above is a useful default.
I used this scene setup, with this base render, and this depth pass. Simplest figures (Poser 4 Lo Res) and simplest setup (a few plane primitives)
Both of the following images were from those same two simple renders.
First image has... background graveyard at night AND old man running AND ghoul chasing (a few more details but you get the idea)
Second image has... background park path daytime cherry blossom trees AND young woman jogger AND middle aged runner (a few more details but you get the idea)
The important part is that the extension more control over subparts of the image
I returned to this because I just watched 'Wag the Dog' again. Love the scene where they have Kirsten Dunst run toward the camera holding a bag of chips and then they change out the background, and change the bag of chips to be a kitten.
These results are amazing. Very clever to first set up a basic composition and then use it as a basis to generate a realistic AI image.
I am nervous, however. I can see the time is coming when a lot of us designers and illustrators will be out of work at the rate AI imagery is taking over. Any user can now generate or "retouch" an image in the blink of an eye, without the education or experience that used to be required to accomplish these things.
All reasonable concerns, @cobusp. On the bright side, my limited experience with AI makes me believe that people who know the fundamentals and are experienced artists will still be leaps and bounds superior to people like me. For example, the AI can be trained to understand the physics of lens concepts in professional photography, and so someone like me doing a portrait and then a landscape will still be inferior to the photography experts. I suspect the same is true for people who have a deep understanding for various painting materials, techniques, and styles.
Poser 4 Dork and his son Poser 4 Boy.
P4 Casual man, P4 Child, a low res house from the Carrara native content browser, a plane, and a Carrara terrain.
The key experiment here is inverting the Carrara depth pass in an image editor then using the result as is in the Controlnet Depth option. Set preprocessing model to none. I also used Controlnet Canny but set to a very low weight (0.2 or so). I rendered the base scene with the Carrara native Toon filter to make it easier for the Canny processor to find the edges. I used inpainting to age the face of the man and add a beard and glasses.