I installed autimatic 1111 and now i wish my gfx card had 64Gb of memory

AI is king, long live the king

^AI vector drawing

@joojaa Pfff, don't whine. I'm living with 8GB vram.

@joojaa Nice! How did you make this?

@joojaa, I asked ChatGPT to write me an SVG image of Harrison Ford in realistic style with gradient shading.

Got this.

@Wrzlprmft, I see you moved a question I answered to Stack Overflow (stackoverflow.com/q/75847228). Fine by me, but I'm just wondering if the css tag and similar tags should just be removed from GDSE altogether? I think that question lived up to the description of the tag: "Ask anything related to implementing styles, effects or graphic-design-specific tasks."

@Wolff using automatic1111 with dreamshaper V4

@joojaa But how did you get a vector output?

I vectorized it with a ai postprocessor

@joojaa Ah I see. Which one?

@Wolff has no name it scalled vectorizer 0.01 on my disk, its probably, a bit worse that adobes tracer at the moment

currently working on making a slide set called is AI going to eat you.

Thats a slide background, still needs some robots on the background. and dimming for text on top

@joojaa Nice. At the moment, making babes is easy, but lots of other subjects are way harder. I can imagine future models being way better with better training images that are tagged better.

@joojaa Well is AI going to eat me?

Yeah because im using Rev animated which makes humans want it or not

@Wolff no a dude/dudette with help of an AI will eat your lunch

@joojaa The Keith Parkinson is strong in that backgorund

@joojaa Yeah, unless I'm using AI myself.

@Vincent yeah weell its a animation isnpired model that makes animationstyle backgrounds

hard part was for me to get it remove the human subject form center without generating 2 humans

Parkinson was a D&D /fantasy illustrator in the 70s and 80s :)

@Vincent i know

and modern animation is heavily influenced

@joojaa Because of the resolution perhaps? It's a quite big image. Did you create it in one go or upscale?

No this is not yet upscaled it is double ai rendered though

you can see the parameters in the file

@joojaa "Upscale fix" then

yeah, stil going to 4x upscaler

its about attempt 14

It's normal to get multiple subjects if you generate larger than the size the model was trained on.

yes

But its not neccesery

@joojaa Ah, thát kind of animation

Lately I've been having fun using ControlNet to flesh out line drawings.

Like this

Or this maze.

Going the other way has been harder. I mean from photo to vector art. But perhaps more useful.

Yeah, the thing is if you run automati1111 on your own computer you have no deecensy filters in place and the thing keeps generating porn

@joojaa It can easily be avoided with negative prompts in my experience.

Yeah but its till tends to show way too much female skin

You can trick the machine with all sorts of prompt hacks.

anyway worth experimenting IMHO

For example writing young sexy woman in the prompt will most likely give you something bordering nsfw. If you instead write middle-aged woman with wrinkles in the face you will get someone who looks 150 years old. But writing something like (young:0.5) (sexy:0.2) woman with (wrinkles in the face:0.3) might level things out and give you someone who looks like they are 40 and sfw. It's chaotic.

@Wolff anyway im trying to do images that have a very standard AI look and feel to drive the point across

Also found that adding something like (zombie:0.2) to the prompt gives people slightly more tired faces that looks less doll-like.

not using sexy though

@joojaa I wouldn't for most things I make, but the funny thing is that it counters wrinkles so having both can keep things balanced.

I laughed hard was generating images borderlivbe sexy and upressed with latent space (now i use sharp) anyway the upscaler turned the image full of female breasts

Like chair was a mass of breasts, legs ended in breasts etc..

Also found that using something like mechanic can be used to make less sexy images without them necessarily looking like a mechanic.

@joojaa You could try another model haha.

@Wolff yeah i did

@joojaa, I also had some fun with giving single words way too much weight. Sort of makes the model go crazy.

Seems like a artistic thing to go crazy

Like this one which contains (palm tree:2)

how do you make inpaint like just inpaint

@joojaa With great difficulty! I haven't had much luck with inpainting really. Or I haven't had enough patience.

Another example. This one has (grinning diabolically:1.3)

And this one has (grinning diabolically:2)

I like to make the machine break. The funny thing is that with "overweight" words, it's no longer deterministic! I get slightly different results every time I generate with the same seed.

@joojaa My main problem with inpainting is what prompt to use. If I just reuse the original one, it doesn't paint what I want and if I only write what is missing I get the wrong style.

ahh thats why having a more art limited model helps

@joojaa I guess so. I tend to be focused on get a very specific style by using sometimes pretty contradicting terms. Can give nice results, but hard to inpaint. Maybe I should keep the style generic until I have the composition I want and then apply the wanted style using ControlNet in the end.

yeah im trying not to use controll net

@joojaa But it's so fun!