- This topic has 193 replies, 60 voices, and was last updated 8 months, 1 week ago by
.
-
Posts
-
@brent that is my take too.
🙂
@eelgoo he actually specified "A New Zealand female business owner..." so I'm impressed it ignored the incorrect "his" afterwards. I assume it considers adjectives > possesive pronouns amongst other things.
Misunderstanding. I didn't make the image and wouldn't know how to. Under the image you can see that I'm asking what I would need to create it. Since then I've watched a lot of tutes and stuff and I'm still not sure if it's worth the effort.
This video was helpful but I know it's only the tip of the learning-curve iceberg.
I read today the AI will kill the stock photography business and I reckon that's probably right.I saw your comment after the image
That’s quite a mouthful LOL. I’d just like to know what I would need to make images offline, what to DL to do that and not pay for anything.
If you are on the Discord, Randy, Ethiopia and I would me more than happy to guide you end help you start your own images OFFLINE. There are a few things to install and download, but after that, you're setup.
Don't hesitate to ping us on Discord, there are channels dedicated to AI and Stable Diffusion. It's more practical and interactive than in here for that kind of things.
Cheers
Math
Thanks mate. What do you think about the video I linked. Does it pretty much cover the installation process for SD?
I read today the AI will kill the stock photography business and I reckon that’s probably right.
I'm okay with that, as long as I can watch the Board of Directors at Getty Images choke on their own blood and bile. AI (as it is marketed and shown today, and not actual Turing-Test-proven AI) will get there - it still has a few snarls to iron out, but it will get there insofar as stock photography is concerned. I give it 6 months to a year before it's on par (or at least good enough for commercial still photography and illustration work). Add a year or two, and we can start putting them into commercials and other short 'animations'. Four years after that, we can put actors out of work (...and with the current crop of spoiled-rotten A-listers, none too soon.)
As far as hobbyist stuff? I don't see AI taking that over quite yet. I love and prefer the control of the CG interface. Telling an AI engine to whip something up, no matter how specific, lacks that element of interactivity for me. It's as much the journey as it is the destination, after all.
So True @Ethiopia. The enthusiasts still buy it and more vendors are providing content, so I can't say it's a failure either (at least, not compared to the new Poser).
While DS is in desperate need of a new iteration, as features and capabilities lag far behind other rendering competitors, that probably isn't happening anytime soon (at least, their job openings don't give that impression). It'll just take time before nearly all new releases are G9 related, to begin filling out their library.
Personally, I barely use it... if at all. I may even jump ship from DS entirely, as I've been eyeing UE5 and Blender, once I upgrade to a better PC.
I do wish they had a Skin Builder for GS9 (or at least 8.1), it's ridiculous they don't so far. Other than that, it's pretty "meh".
EDIT: Also, as sexy a pair they are... Stable Diffusion or Midjourney combo'd with ChatGBT, there still is that thrill from creating a render by hand: Setting up the scene, polishing the textures/shaders/models to how you want them, the lighting, the focal distance, this and that. But I do agree that they are pretty useful and amazing. I wish DS could do something to keep up.
I watched the video you linked and it's accurate, apart from the Stable Diffusion version that you should download: currently 1.5 and 2.1 are out and it's a good thing to have them around as more and more checkpoints are built on those (even if you won't use them "vanilla"). 1.4 is still around though.
I usually go to
- Sebastian Kamph
- or this channel Aitrepreneur
- or this one Olivio Sarikas
They all have great tutorials and the first and last one explain really clearly what Loras or Lycoris or Hypernetworks are, plus technics of all kinds
I'm sure, as a DAZ afficionados, you already have a decent graphic card (I'm too lazy to climb the stairs and go in my home office to use my main rig so most of the time I use my laptop with an NVidia RX 3050 Ti and 4 Gb of VRAM with no issues apart from image size limitations), but be advised that, as with DAZ, HD space could be an issue: I have close to 100 different checkpoints and Loras and the install folder is already close to 200Gb in size 🙂
Generated images will also soon pile up and eat up some serious space 🙂
You'll see that after a few hours of adaptation, you'll feel comfortable enough to start fiddling round with the parameters and you'll soon start downloading other networks and Loras.
Model is a "generic" word that covers the checkpoints (huge files containing training data), Loras that are similar to checkpoints but trained on a particular subject, textual inversions (called embeds) and all the rest.Two good sources to downloads "models":
- Civitai, very nice site but sometimes overloaded and experiencing access issues (work in progress, victim of it's success)
- Hugging Face, absolutely ugly interface and messy, but huge source
Cheers
Math
Thanks @Mathmathou and sorry everyone else for going OT on this thread. Last weekend I bought a new 2TB HD so storage won't be a problem for a while for the SD files, they are really big. I guess like 3D I need to experiment with SD to learn how to use it. ATM it may as well be magic as I know next to nothing about it or the terms you used (networks & Loras). Maybe soon I'll bite the bullet and set up SD but for the moment I'll do some reading and watch a lot of tutes. Thank you.
Still installing mathmathou. Got hung up with getting GitBash to recognise/accept c\ai folder but eventually worked. Then I had to edit the SD webui-user.bat file to show it where python is installed, that was more fun not. It's now installing (another) 2.6GB of stuff. Hope it's worth it.
It is!! You'll love it.
Your issue with Python: did you check the box stating to "add Python to Windows paths"?
Since you are editing you webui-user.bat file, you might want to modify it a bit more and add
--xformers
just after
set COMMANDLINE_ARGS=
to obtain
set COMMANDLINE_ARGS=--xformersThis argument will optimize your VRAM usage.
Some WebUI that use the Stable Diffusion "engine" are installer with PyTorch v2 (like VladDiffusion for example). PyTorch v2 optimizes the VRAM usage enabling crazy resolutions of images in creations and upscaling.
Automatic1111 comes with an older version of PyTorch and, until they update it and make it fully compatible with PyTorch 2, the --xformers argument will help you.
Is that too much info? 🙂
Thanks @Mathmathou. Everything work fine now. Oh boy SD is addictive, been stuck in my chair for the past 3 hours.
Sorry to ping you again
Last minute "must watch video" :
AMAZING SD Models - And how to get the MOST out of them!This video just came out and, for a beginner (no offense meant), I find it very instructive. It's not about installing Automatic1111, it's about adding Loras, why and how, same for VAE, models and embeddings, useful settings and a few tips and tricks.
There is a comparison between 3 highly rated models and I found it very instructive.
Have fun 🙂
Math
Last minute edit: I was reading the release notes about the last Stable Diffusion version (released 2 days ago) and they mention that they switched to PyTorch 2.0. So if you have an NVidia GPU, I think you don't need that --xformers addition to your bat file.
Did you install NVidia CUDA SDK by the way ?This conversation is very interesting, I take notes... I will probably make the jump at some stage even if I have strong "philosophical" and technical doubts about AI art (if only it could maintain constant characters). However I am frightened by the HD space cost (mine is so cramped because of DS and MSFS).... But many thanks to the contributors.
- You must be logged in to reply to this topic.
