Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Very cool, @Diomede! I've done a lot of img2img and controlNet processing in Stable Diffusion as well. Tons of fun!
The changes may be subtle, but they're a definite improvement. The eyes, eyebrows, hair and shadows on the nose definitely look better, and most importantly, the likeness is still there. Great job, @3Diva!
- Greg
@3Diva - subtle but significant. Great job! I had to look very closely to see some of the elements like the shadow on the eye orbital and similar. For people going for photorealism, that sort of subtlety is a hude deal. Congrats.
Thank you for the comments, everyone.
Here is one more from, probably the last for a while. Here I was going for a so-called 'Bodice Ripper' cover for romance novels. I had a very hard time getting the male character to face away from the camera. My toon to canny guide faced away, my depth pass faced away, my text prompts said to face away - just a huge resistance. Here is what I finally came up with, along with the 3D elements. Loosely based (Very Loosely!) on a cover for Catherine Coulter's 'Devil's Embrace.' I think the cover artist was Elaine Gignilliat, but if not that specific one there are plenty of other covers by Elaine Gignilliat that are close enough.
https://sweetsavageflame.com/romance-clinch-covers/artist-romance-covers/elaine-gignilliat/
This is another example of pure text2image with help from Controlnet, not any image2image.
Mistress of the Seven Seas
Result - Mistress of the Seven Seas
click video
click video
I did similar thing to Diomede for a video
Looks interesting, Wendy.
I am still fighting with ComfyUI to get similar functionality like with Automatic1111.
Below is my attempt to expand the image 200 pixels down.
@algovincian Thank you! The subtle changes can be nice. I think Iray can already push the realism envelope pretty far. It would be really cool if we could somehow integrate Stable Diffusion into Daz Studio. Allowing it to take over after Iray does it's thing. I'm sure I'm only dreaming though. lol
ROFL Now I can't UNSEE IT! You're totally right - it does look like her!
Thank you! I agree, that even a subtle change can make a nice improvement and push it that much farther into realism. Iray gets it like 90% there, would be nice to have Stable Diffusion inside Daz Studio to push it that extra 10% into full realism.
The art you're working on looks awesome - I could totally see these as book covers on romance novels! I love it! <3
Video
some AI backgrounds, filters, animation of rendered backgrounds
the music is AI https://aitestkitchen.withgoogle.com/
Here is another image against of the normal workflow people are normally doing.
1. I started with an AI-generated image but as commonly occurs the eyes are a bit messed up.
2. As needed it for a really high-resolution banner I started generating several DAZ renders 6000x6000px, trying to match the light's position (Not the skin tone at this stage).
3. Originally I replaced only the eye.
4. But also I wanted texture on the skin because the AI-generated images are a bit flat. I used the Frequency Separation to integrate the texture to the existing image.
5. Then added the mouth and the second eye.
6. And the shoulders to maintain the proportion. The hair and the banners are AI and it has some vector shapes. All the project was a bit challenging because it was complicated to match the fisionomy of the face.
Thanks for the comments.
I really like how you fixed the eyes, @Rafael. I have occasionally gone from AI back into 3D to add some props and similar details (using shadow catchers, for example), but have never tried something integral like the eyes. Fantastic job. Thanks for sharing the info.
Agree, looks great. I would not even think about changing the eyes.
Been about two weeks, so I am wondering (a) how much I have forgotten, and (b) how many new transformative tools have been released that I have not heard of. Here I am dipping my toes back in the water. I used the 'Latent Couple' extension of Stable Diffusion to divide the canvas into separate regions so I can instruct a horse on the left, a Canadian Mountie in the center, and a fugitive in the right, and attempt to keep the prompts from affecting other areas. The role of a couple 'blockout' renders is essentially the same as others I have posted in this thread. If anyone is interested, I have posted a more detailed workflow elswhere. https://www.daz3d.com/forums/discussion/comment/8645931/#Comment_8645931
Great work, Diomede, and thank you for posting the link to your workflow.
not my art remixed though definitely AI
a video I made using several sites mostly Leonardo AI
Looks very nice, Wendy.
I didn't even create the prompts
I just typed simple stuff like Robin Hood, Maid Marian etc and let Leonardo generate some
make Art button st it's finest
Wendy, the video elements are definitely getting more consistent.
I caught the end of a webinar in which one of the participants was using ChatGPT to generate the prompts for one of the AI image generators. It helped generate some of the more technical photography specifications for the lens, etc.
Windmill pinup
One more demonstration of low res models to block out a scene. Will post side by side with workflow in next post. The following could be done with Blender or.... Or, Daz Studio and Poser users could adjust the figures they have. I modeled and rigged a low res beauty suitable for toon renders using Carrara. The emphasis on toon here is to aid the 'canny' preprocessor in the AI program to detect the edges. Similarly, the terrain and the windmill are low res models created in Carrara. In this case, although the blockout elements were designed for toon, the final output is more photorealistic. I generally prefer pulp, illustration, and toonish stuff, but this is a demonstration. Here is the final result.
And here is the workflow.
Thought I would join in now I have Krita AI up and running
Redhead
Vicky 4
Very nice renders, everyone.
Love to see other people doing this sort of thing, @Diomede! Great work.
Hard to believe it was 9 years ago when I first posted here about NPR and what I refer to as "analysis passes":
https://www.daz3d.com/forums/discussion/comment/795474/#Comment_795474
A lot has changed since then, but many of the concepts from back then are still valid today. I strongly believe DS is uniquely positioned to work well in conjunction with all the new generative AI tools.
Anyway, here's a quick video showing the output from a recent workflow applied to 4 different characters. In this test, 4 different subtle expression changes were made from one single input image for each character:
- Greg
@algovincian, thanks so much for the video, the link, and the comment. Great info. Going to be a big help to my efforts right away. Much appreciated. And in the longer term, I had started to put together a screenshot walk-through of using Daz Studio canvases and the bridge to Hexagon the way I have been using Carrara. But as the phrase 'had started' implies, I didn't finish it... yet. Your video and link will be great resources when I decide to finish.
Here is another recent effort. The AI database (model) I was using had a very hard time with the concept of stegasaurus, so this isn't what it should be. But in making it, I did use the 'shadow catcher' method to reload an early Stable Diffusion output back in my 3D program as a backdrop, then render again. Just about any 3D software should be able to do a similar approach. People interested in the workflow, can find the first post of a walk-through series here. https://www.daz3d.com/forums/discussion/comment/8654701/#Comment_8654701
For your information - there is this.
https://www.daz3d.com/forums/categories/daz-ai-studio
Daz3D AI generation new tool test
Visitor