- This topic has 58 replies, 14 voices, and was last updated 2 years, 7 months ago by
.
-
Posts
-
Here's a point from an older conversation we had a couple(?) of years ago.
I mentioned that because I was on a more modest PC, I didn't just make scenes with the goal of making a render. It would be set to render overnight or when I was finished (or given up) and was doing other stuff on my online computer (an older PC). What I was mostly doing was playing with digital dolls. I'd load a character (I have a bunch of V4 based characters), pick a setting, then set about adding items and dressing my character. I could easily spend an enjoyable evening trying out different clothes, items and poses. At bedtime I'd set it to render. As I goofed around I'd take screenshots. The scene and screenshots where all saved in one directory prior to rendering. I have hundreds of these all saved and frequently revisit one. I found that a few others had the same workflow too.
Even though I now have a much better graphics PC with a good card (3060 12GB), I'm aiming at doing image2image with the intent of improving the screenshots, and I'm hoping that it will do what Studio is so lousy at and AI is so good at, draping cloth realistically and displaying realistic fabrics like Satin and sheers.@Ethiopia "Several hours for a decent image" isn't render time, I can spit them out in seconds. It means all the time spent stuffing around with mixed loras in multi-characters scenes. It's 10 times as hard as you now have 2-3 people to perfect, and when one happens to look good the other/s don't and vice versa. Regional Prompter helps with image composition but isn't that reliable. For me it works like this-
1 person = easy
2 people = very difficult
2 people interacting = extremely difficult
3 people = good luck with that shit@frank22 I'm running into those same issues a lot. DAZ can be helpful in that scenario a lot of times for me, I can generate a very basic scene in DAZ and use that in control panel of SD for one of the references, or cammy etc. Or even in img2img. It can help form the basis of what you are trying to do and feed it to SD
Thanks gts6. I've been playing around with DWPose with controlnet where you can use a photo to pose your character/s. It kind of works but like everything else with generative AI it's quite unpredictable.
I've seen some impressive stuff on Tumblr but no one will share their secrets. someone will eventually, but for now many people want you to buy their renders on Patreon or that fan thing.
Nice examples @Legolas18 but toons are much more forgiving than realistic likenesses of 2-3 real people of different gender. I've seen some excellent SD renders of this but can't easily pull it off myself without a lot of trial and error = hours 😢.
@ethiopia I suspect many people use Midjourney, which gives really impressive results, but follows prompts quite poorly.
On the other hand, I prefer Stable Diffusion with the multitude of free expansions found on CivitAI.com (just apply the search filters to see only what you want). Most images even have an (i) icon where you can see the generation data.
@Frank22 Now I understand why it takes you all so long to do 😢 , but that's why I prefer toons, much more colorful and it's easier to fix various mistakes in photoshop.I'm happy with DAZ3D right at the moment. I looked at Stable Diffusion a few months ago. Maybe sometime in the future I might use it. From what I have read, it seems SD can't handle where there are multiple models. DAZ can.
Just watched a video, not sure what I exactly saw, but I am impressed.
I guess it is AI as well, but integrated in software I might use. Create figures in DAZ, export to...
linkYeh gaver, everyone was talking about MetaHuman a year ago but it kind of died in the arse.
This is all the rage now.I have seen that to, also pretty amazing...But it is still prompt generated ( as I understand). Like more people here and on discord, the main hiccup with AI is that we have to little control. I would love to import chars somewhere , like with this meta human thing. Maybe it is just that I am old fashioned and used to timelines and sliders...
you can at least create good facial expressions this way I guess...
The things DAZ produces aren't all that fantastic@legolas18
Just my contribution to the three bodies problem, but in Daz Studio....
So the story of three girls in the forest...
Some work but less than 2 minutes rendering (thanks to really powerful rig). The original is in 3840*2160 if you are interested.
What did they see ? A wolf , an orc or ...
I do not think an AI can currently provide a good and consistent answer. Daz remains useful (but for how long ?)
- You must be logged in to reply to this topic.
