- This topic has 39 replies, 15 voices, and was last updated 2 years, 11 months ago by
.
-
Posts
-
welcome back @frank22. I think @mathmathou has grossly exaggerated my capabilities with SD but certainly the ability to combine Daz and SD gives almost limitless possibilities with inclusion of some of the extensions like Controlnet - whether it is taking a pose from a render or depth map from an environment you can set some basic parameters for what your image will end up looking like. Would also recommend this channel https://www.youtube.com/@AIchemywithXerophayze-jt1gg. Although he is predominantly focused on monetising his Chatgpt prompt guide (which i have found to be very good) he does explain some things in an easy way to follow
Thanks for the welcome back guys. I did my first DS renders in 2.5 months yesterday.... I have to admit I kind of missed it 🤗.
@frank22 Welcome back.
Have only dabbled with online versions of SD and MJ ports (eg OpenJourney). Even tried various extensive prompt and negative prompt suggestions to try to get to more specific and cleaner results, but man is it a sensitive art/science. Still not at all satisfied with the results, especially relative to all the incredibly realistic creations out there. (Still learning.)That's text to image, obviously. The prospect of image to image, or image enhancement, from an original DS render is intriguing, but haven't tried it yet. Seems it would be far more accurate with far less or fewer extra limbs or anomalies. And no or less need of postwork or filters or LUTs or things. Guess it all comes down to original intent vs. final output.
None of it usable in contests, of course, but very interesting experimentally. As another huge fan of ultra-realism, seems there isn't much better an avenue when trying to use DS. (For many of us who feel like we're constantly hitting the "DAZ Ceiling", maybe the only thing that allows to compete with other, spendy render engines for more realistic output.)
[Favoriting this for links and pointers. If there's ever time to use them.]EDIT: Should someone start a non-competition AI-Enhanced-DS-Render-Demo thread? Would it be condoned in the Art Comp forum with the right guidelines and statements? Or would Art Showcase still be more appropriate? (Know it's a sin to some, even an "abomination", but knowing things are separated from original DS works, seems innocuous enough for artistic purposes either way IMHO provided of course the qualified statements and understandings.)
I learned today that increasing the sampling steps can greatly increase image clarity. Copy/pasting the generation data from images at Civitai is a good first step to learn prompting.
My workflow is basically to render at 512x768 (or vice versa) until I get something I like, then use hi-rex fix with the 4x-UltraSharp upscaler, then increase the size until I run out of VRAM. I try to avoid sending it to the Extras tab as it kills the details....looks airbrushed. With 8gb of VRAM I can get up to about 1600x1800 image size.Good to hear that you are okay.
I had a look at AI to make art, but lost interest. Might look at it sometime in the future - but happy with daz at the moment.
Yep there's nothing wrong with DS and I'll continue to use it. SD is just another tool to play with. It's all cool 🥰.
Ah He's back!
SD images look nice, so thought I give it a go, followed the steps Mathmathou provided. Didn't get past instal Python. After that run the script which says there is no Python. Was late at night 😉
But I guess I will not be using SD, I like to create as I go, so no finished pic in mind. Just set up a scene, look for nice props that fit in, fiddle about with lights etc.
I render just enough to see what effect I created. If not good, stop render, adjust, try again.
It is the journey that matters, not the destination. But that is how I see things...So true! My wife has a sign saying exactly that in the kitchen: "It's the journey that counts, not the destination".... Have you been in my kitchen while I was away by any chance? 😉
If it was late at night, I bet you omitted to check the "add python to path" when installing it. That's my first suspect on the list of probable causes for the "no python" error message.
Cheers
Math
Do you use Kohya for Lora training Mathmathou? I tried to get it running but without success. The GUI won't load up for some reason. A python issue but I already have it installed for SD. I think I know what I did wrong..... maybe.
@Abad LOL
Yeh Abad but sometimes we come back 😵.
dirty naked zombie walking in spooky cemetery, full moon, night, fog, nsfw, <lora:add_detail:1>, masterpiece, best quality, bluish lighting, ghastly,
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 286469769, Size: 768x768, Model hash: dde3b17c05, Model: aZovyaPhotoreal_v2, Denoising strength: 0.37, Hires upscale: 1.5, Hires upscaler: 4x-UltraSharpTime taken: 43.99sTorch active/reserved: 5054/5892 MiB, Sys VRAM: 7885/8192 MiB (96.25%)
I've been going through Civitai and grabbing stuff I'll use. I hate that it's an endless scroll, I have no idea how far down I am. I may be at it for weeks. The search sucks mostly because few use appropriate tags and everyone uses their own language.
And why is everyone so hung up on that Emma chik?Ha ha I think there are now at least dozen Emma Watson Lora/TIs at Civitai and I suspect not everyone is treating her respectfully 😢 considering that about half the site is pron related. You can also tell that much of the user-base is Japanese/Chinese... tones of Anime and Asian related stuff, nothing wrong with that.
I'd like to be able to search from oldest to newest but you can't do that. Anyway, at my age it's probably healthy to learn something new from scratch to stave off ever-approaching dementia 😁.
- You must be logged in to reply to this topic.
