Posts

@tyto4tme4l
@Lord Waite
Yep.

Mind, this isn’t specific to AI, I’ve made loads of vectors where I notice something’s wrong or missing after I considered it “done”, and have heard the same from more “traditional” digital artists.

@tyto4tme4l
Definitely. Happens embarrassingly often when I’m in the middle of tagging something before I hit submit. Often I was distracted by something else in the picture I really liked, or it snuck in while I was changing something else…

@Background Pony #D936
Noticing a glaring flaw once you think you’re finished.

@Teaspoon
what is the most frustrating thing about the process of getting a good ai image?

@Background Pony #D936
Character aware prompting, or some form of smart regional prompting. Ala “Character A has red eyes and yellow hair, character B has blue eyes and red hair, character A is dancing”.

Would just make life easier…

(there’s much more impressive stuff going on ofc)

so what is the most impressive ai development you are excited about?

@Scarlet Ribbon
That’d be nice! I would appreciate it if you’re willing to share with us. I’m always looking for ways to improve my workflow and learn new techniques. Though I’m not in the Discord I’ve seen mentioned so I can only assume that’s where most of that sharing happens.

I did have a question about that. I’m not an artist, but I did wonder if it’s valid for me to generate images to trace. It’d be a step up for me to be tracing something instead of not drawing at all.

@Star Streaker
I’m rebuilding it right now, but this new version is designed to potentially be distributed and is less dependent on Wildcards in favor of ComfyUI nodes that handle everything in the workflow directly. The only things that would be needed would be the various LoRAs that are invoked, and I’m making a toggle for each one of those, with clearly labelled notes that link to the needed LoRA downloads.

@Scarlet Ribbon
That just makes me want to see your workflow even more… Mine are pretty basic, all things considered. I let SwarmUI handle the wildcard stuff.

@Background Pony #D936
Difficult to say. AI-generated backgrounds could be detected, but using AI-generated images as a reference or a base for tracing can’t really be detected, as far as I know.
More popular artists wouldn’t even need a LoRA for their style. Models like ZoinksNoob can already mimic a lot of artists’ styles to some extent.

It would be especially tempting if a lora for your art style already existed, I’d imagine.

(And I’m sure I saw a comment once somewhere where an artist had decided to go all in on AI art and personally trained a lora on their own art style. Not a pony artist, though.)

@tyto4tme4l
what do you think would be some tells? i don’t want to put anyone on the spot by naming names but i am curious as to what might be signs of ai usage

@Background Pony #D936
Oh yeah, totally. The temptation is too strong.

do you think there are pony artists that are using AI and hiding it?

Or maybe they just want the humanoids to be on same difficulty level as generating ponies. They are just so easy to generate without any major flaws, unlike other species.

I have a theory that MLP ponies are biomutants, genetically engineered specifically to withstand the corruption of rawgens.

@Denatural
oh thanks for that. interesting point on the hands. i guess they are trying to keep in mind mistakes an artist might make?

I can do you one better. here is quality guideline https://derpibooru.org/pages/aiquality
And they are still pretty lax on the hands abominations.

@Denatural
can you link to an example of the sort of ai image that derpibooru still allows?

Considering derpibooru still allows AI with “significant edits and direction from a human”,this site will not do well.
All good AI pictures require significant edits and lots of img2img to look good, so this site will just end up as a burial ground for all the AI slop.

You misunderstand. ComfyUI is what I use, and ComfyUI doesn’t just attached the prompt to an image as metadata like, for example, A1111 does. It attaches your entire workflow. There is no meaningful information someone can get from my workflow (except, possibly, that I am insane), to recreate anything about an image I make, as nothing in the workflow would contain the prompt that was actually passed to the sampler to create the image. I literally don’t even know what the prompts are myself, as that information isn’t set to be saved, I just get an output image.

I could technically alter the workflow to save the prompt as a separate file with each generation, but that would just create a mess of files to clean up.

I see what your saying but I still maintain that prompts should be with the people, I wouldn’t need your workflow, rather there’s a base image I like and I re write the prompt from that. I learn a lot of things from working from others base prompts.

@mp40
I use a complex workflow that relies on a mess of wildcards that wouldn’t work for anyone, so I try to strip my metadata whenever I remember to so I don’t have to explain to people why it doesn’t work. That said, a lot of people are willing to share their prompting tricks and stuff if you just ask on Discord.

@Hakirayleigh
Obviously. If you don’t see them, maybe you should check your filter.

Is this site allowing NSFW images??