- This topic has 17 replies, 10 voices, and was last updated 1 year, 5 months ago by
.
-
Posts
-
Given that yesterday the new generation of Stable Diffusion (SDXL 1.0) was released, I wanted to showcase something I made with it (granted I still have to learn how to make the best use of it).
Also, not all extensions work yet, but I am really impressed by what this new model can do.
This image has also been edited quite a bit to fix all the problems that AI has with images as well as some color toning in Photoshop. Enjoy!
@legolas18 Presuming that this was done with prompts only (that is, from "scratch" with positive and negative prompts and text-to-image), SD does have an "enhance" mode as well, correct? (As in, image-to-enhanced.)
In one of the other AI/SD threads there was talk or suggestion of taking original DS renders and AI-enhancing them. I've yet to try but was considering with some of past work just for fun and testing. Have also only tried text-to-image with online-hosted Github style pages, and not yet downloaded anything locally. Guessing the results are considerably different.
@spats This is not just text to image. That's just the first step. After that, I did an img2img as well to add details, plus manual editing in Photoshop to fix mistakes (for example, I painted the eye myself) and enhance contrast and colors.
But yes, what you are talking about (img2img) can yield good results with renders, depending on the pose of the character.
@legolas18 Ah, interesting. So a two step (or more) process depending on what you're after, the cores of which are still SD. Image-to-image is intriguing to me; especially if the prospects of keeping as much of an original render are possible depending on settings. For example, I'd be interested in not much more than taking an original DS render, keeping lighting tone and style as true as possible, and getting an Arnold level sharpness and realism out of it via SD. (Just realizing DS and SD are semordnilaps.)
Keep up the beautiful and interesting work.
P.S.: What's your favorite source for promts or prompt tutorials? [EDIT: That question goes out to the room.]For sure, it is an very nice image and a good work from AI and You @Legolas18
one little comment, the book looks too heavy in her small hand, she holds it as it is just a single page 😉@spats I forgot to mention the inpaint function 🙁
You basically "paint" over a small section of the image, and the AI recreates that part. I usually use it when some parts could use more details, or I wish they were a little different.
As for prompts, I go to Civitai.com where you can download for free all the various add-on models. There, they have a lot of images and if you press the (i) on the bottom you can see copy/paste all their settings and prompts. Just choose a style you like, and start experimenting until you get what you want.
Also, remember you can add to your prompt an artist name, and it will change the image to that style! This image was made "in the style of Sophie Anderson", for example. for more styles take a look here:
I love it, but I do have a critique. The book is as big as her torso, which means it will be heavy. Look at how she is holding it one-handed, from a corner. The image is great but that one thing makes my brain hurt.
Thanks for pointing it out, it is sometimes hard to see all the things the AI does wrong.
I now made the book smaller with a little bit of Photoshop and some AI blending. Enjoy 🙂
I wouldn't have bothered to do that Legolas. Just say the book is made of vellum pages, which it would have been in old times, so it's very light 😊.
@frank22 Thanks for letting me know, but I got biased on how books look today, so maybe more people are familiar with that kind of look 🙂
My thoughts. Stupid things like gravity don't apply when the word magic is applied.
- You must be logged in to reply to this topic.
