- This topic has 27 replies, 7 voices, and was last updated 1 year, 3 months ago by
.
-
Posts
-
Well I'm still a bit concerned whether of not the images fall within the ZGFX TOS. I actually re-renderd the initial two images because the anime AI model I used produced faces that appeared too young to be safe. You can't be too safe these days with regard to that sort of stuff or someone will... whatever.
Glad you liked it Adelsia. My memories of stuffed trains on wet mid-summer nights on the Ueno/Hibiya line has have left me somewhat corrupted, so much close sweaty bodily contact, so many beautiful women, but not a chikan mind you, I would never do that.Note: a Japanese woman would never think about going out without a bra. Reality and perverse desire at this conjuncture depart from one another in the "art" world. I make no apologies.
I like it
it is great to be able to render such a crowded scene.
If it's music, this comes to mind....
I think people dislike AI art because they might have seen some of the first attempts at it (which were a bit...atrocious) or because of their "AI" look.
But everything has its own look? 🙂
Anyways kudos for the pics!
I think the biggest objection / hurdle at present is the inability to get consistent similar images.
Difficult-to-impossible to create a scene and then only be able to change one [for example] object in the scene.
At least with a 3D render you can revisit it and keep as much of the original identical whilst adding / changing any of the other content.Things change so quickly - looking forward to a true blending of the two formats, to really push creativity.
Also, for beginners, the lack of free options to experiment, as almost all [I think] of the leading players are now using a pay-as-you-go access.
@Fuzzy3000 I made an entry about render to AI here:
So, technically you could use AI as a render engine 😉
Also, there is a free option, it's called Stable Diffusion (which is a leading player). You install a UI on your computer, download one or more free models...and you can create as many images as you want 100% all for free.
As for making something like a Graphic novel, I guess it's not possible yet, but for now you can only get consistent faces.
"All women are wonders, because they reduce all men to the obvious." Name that quote🤭.
@Legolas18
I need to do a lot more digging to get better acquainted with AI.
Does SD have limits on what you can do for free?Most that i have checked seem to try to entice you in and then cut you off unless you subscribe.
At present I can't see myself justifying a monthly cost for what would be limited use.
And, as mentioned, I can re-visit a render and change anything - pose, expression, clothes etc. whilst leaving everything else as original.I did find one site that seems promising as a free introduction:
Unstability.AI:
You can use the site for free. However, you will be limited to 100 slow credits per day. If you would like more than that, or to access fast credits, you will need to purchase a plan.
You can sign in with either Discord or Google accounts.Just tried it with a very basic prompt;
slim blonde female, dressed like Harley Quin, with ripped white t-shirt and hyena pet, photo realistic, in front of destroyed buildings
And the result was quite good.
Still lots to learn... 🙂
@Fuzzy3000 I am not using a website to generate images, rather I generate them directly on my computer with my GPU.
First you need to have a webUI (basically it's a program that let's you do things, but it runs from your web browser). I am currently using Automatic1111 (because it allows you to also install various extensions, and it's the most popular). Also, it's 100% FREE.
The models I download (again, all for 100% free) from civitai.com (these are community made models and enhancements that improve the base model).
NOTE: There are several versions of Stable Diffusion. The most popular was 1.5 until mid July 2023, when SDXL 1.0 was released, which I am now using almost exclusively at this point, because it's really powerful and gives me the best results (all the stuff I am posting here was made with a SDXL model).
To keep up to date with AI (and for guides on how to use things) I follow a few people on YouTube:
https://www.youtube.com/@OlivioSarikas/
https://www.youtube.com/@sebastiankamphMartinis ✔ Paul.
@Legolas18... I haven't moved over to SDXL (yet) because-
* it looks complicated
* all my old stuff (loras/checkpoints/lycoris) won't work in SDXL, and vice versa
* the SDXL checkpoints are massive (>6gb) and I don't have a lot of disk space left
* SDXL v1.0 censors NSFW contentLegolas, can you explain what "gives me the best results" actually means? Speed, picture quality, ease of use, following prompts etc?
1) SDXL is actually more simple than SD1.5, it is quite easier to prompt (although a lot of people mistakenly still prompt it as 1.5) and it requires a lot less negative prompting. Also, it follows color instructions much better, and the hands are improved (but not perfect). The Refiner I only use it rarely (I prefer High-Res Fix).
SDXL also tends to give better results out of the box (at least, some models do, so I just have to keep on testing them to find the best one, which I was actually doing right now). But overall, they give better lines continuity, and with one model you can technically do almost everything. Also, 1.5 tends to have a more "AI feel" which SDXL doesn't seem to have as much. Also, the default image size is 1024x1024 instead of the 512x512 of SD1.5
2) Yes, old LORAS, etc need an XL version to work. But now the civitai community has already created a good amount of them.
3) Well, there are already NSFW models and LORAS. Male "members" also come out infinitely better than 1.5.
4) gives me the best results: Well, certain styles just look wow (I am using an extension where i just click a style and it automatically adds the style to the prompt. Photos and Fantasy stuff look just awesome. Also, overall, I am doing a lot less tinkering on SDXL than SD1.5. Also, the quality and details are better in SDXL. I remember doing a lot of inpainting in SD1.5 to improve quality. I practically almost never do inpaining with SDXL.
If you want to try one model that is giving me really good results (and has NSFW too), I would suggest:
(I prefer the 1.0 version rather than the 1.5 version)
https://civitai.com/models/124419/realmix-xlFor the negative Prompt:
https://civitai.com/models/128969?modelVersionId=141278EXAMPLE: (with High-res fix and Adetailer)
This was made with Realities Edge
Great picture @Fuzzy3000. If you've got 8gb+ of VRAM I'd suggest you should do a local install.
Thanks Legolas that's good to know. I think I would need more VRAM. When I saw all those SDXL reroute nodes I had a panic attack. If everything goes over to SDXL I might be forced to change.
This pic used a DucHaiten style model so it isn't designed to be realistic. The previous pics were a tad sexist so I thought I'd change the theme.
- You must be logged in to reply to this topic.
