Posts

@Background Pony #46E0
A bit of both. In many cases, the way a pony might do a given action is very different from how a human would perform the same action, and as such, many models fail to come up with equine-accurate ways of portraying those actions. IMG2IMG and ControlNet based workflows both can assist to a large degree on that front, and I have used them to considerable success in some outputs, but that requires having a good source image first.

You mean these actions are not applicable to ponies at all (like because they don’t have hands or something) or you mean these actions are too hard to generate with ponies because the model doesn’t understand what to generate? If the second case, can you please give some examples of such actions? I would like to try out if it can be generated with img2img.

@Scarlet Ribbon
I’d say your half right. People are pretty creative, most just do / choose to do NSFW stuff, or so it seems.
Your completely correct as far as trying to figure out what the model understands and doesn’t understand (I recently did a Christmas themed image and found out that the model I was using didn’t understand what horse reins were, had to use a NSFW at very low weights to get it to work).

@Background Pony #A3FD
I’d like to see that too, more interesting SFW type stuff. It’s actually what I’m slowly working on right now. Working around what the model knows and doesn’t etc is definitely a challenge. Been getting help from various discord servers here and there but I’ve only just started down this path.

@Background Pony #A3FD
Part of the problem is that most people aren’t very creative with their prompts. I try to have a fair amount of variety, but there’s a lot of actions that don’t translate to ponies very well, and finding that intersection of ‘the model understands it’ and ‘the prompter thinks to try it’ isn’t always clear.

What I’d like is more images that are not saucy at all. I like happy ponies in particular. The problem is that most of those tend to be simple images of a pony standing or doing something trivial. I want to see more interesting images. I want to see unique styles. I want to see funny comics. Seriously, I can give you some 2-3 panel comic ideas which are at least better than nothing.

Regarding the lack of comments, Derpi has more but it also has more users. Tanta is small. Images usually get like 10-50 upvotes. 2 comments on an image is already a lot.

@truekry
I’m rather shocked that one wasn’t on my faves list on Derpi and had its vote already migrated over. That is classic clop.

You can get more horse like features very easy. Like here.
I didn’t add “My Little Pony” to the prompt and tpony, and offshoot of ponyv6, made this from that.

@truekry
To your point, all my biggest peeves of AI ponies are the same as the popular trends I despise in hand-drawn ponies. I wonder how img2img using IRL horse photos as reference would change things…

[I have an M1 Max with 64GB of RAM, so I’ve got the power to run it locally. Just need to actually get things installed.]

@🐴
The thing is, we have little choice in that regard. The loras for that are all “big mode.” Getting smaller, oval shaped ones, aka more equine in nature, was always a battle fought with the models and loras (for the ponies in general) with a lot of manual editing.

But see, this feedback alone helped. We can talk about issues like that and why things are as they are.

@truekry
Some of it is that listing specific likes and dislikes would get repetitive and spammy for me. There are only so many times I can complain about the giant floppy crotchboobs everyone else here enjoys before I get myself worked up and become the first person to be banned for heterophobia. Likewise, “finally, a horse-shaped horse” has too little variety to be intellectually stimulating.

@Heat Sink
I leave a comment from time to time too, positive or negative. So far only 1 or 2 creators showed that they reacted and 1 lashed out at me for pointing out a mistake. It is what it is. I rather have it that way than total silence.

@truekry
About a month ago i personally did comment a bit on images that were posted here, but i only pointed out the positive stuff of what i liked on a picture. Mostly because i am unsure if the creator of said image would appreciate feedback on their images.

@truekry
I agree on missing feedback. Being told something as simple as “this is a cool idea” or “I like how this came out” would be nice.

@truekry
Hear here!

The worst thing about the silence is the missing feedback in my opinion. What did the people like about a picture, what did they dislike? Why did you get a down vote, why and up vote? It hard to guess. The only thing you can do for sure most people will like is plot to the face. I know, sex sells and all that, but I find it sad that we have this powerful tool that is AI, but if you want people to notice your art it better have plot in it. It also gives tantabus the vibe of a AI pony porn site and less AI pony art site.

@NowIveAwoken
The flood of unedited gens is what got AI art removed from Derpi in the first place.

AI could provide better quality graphics than human art, but it alone doesn’t make generated images more valuable. Especially since AI can generate countless amount of such pictures. There are simply nothing unique, nothing memorable about them. And AI people doesn’t make things better generating dozens of basically same AI artworks.

A meme, a joke or a story incorporated in works of traditional artists is not a compensation for a less quality (it could be so but it’s not the main goal). It’s a thing that make artwork interesting, impressive. Good look on itself rarely make AI image impressive, especially among lots of other relatively good looking AI generated images.

@Background Pony #F22A
Generally speaking, current models are better than most “traditional artists” IMO, but gen AI is also less flexible and often makes some errors, which require human intervention to fix.

in terms of potential is generative AI already better than what “traditional artists” can do? if not how far off are we from that in your estimation?

@tyto4tme4l
Precisely, since the goal IS “anyone can illustrate this”, good quality is an added bonus, and the only “ trivial” problems are getting the people with things to say to realize that, and getting the target audience(s) to adjust their expectations for what “I have little artistic talent, but I do have passion to say something” LOOKS like visually.

This thread and/or site could contribute to both of those with some focused intent, but I have pushed “the ball” as far as I can think of how to do (at least for now), and I hope other people pick it up and run with it at least a bit further here.

@Background Pony #F22A
I would say there is no need to make up for crappy production values with AI-generated images, since it’s trivial to make good looking, generic pictures, while it takes a bit more effort to create something even amateurish-looking with additional elements added manually. It would be more effort for (arguably) worse result.

@Background Pony #F22A
Maybe people just don’t think of using a drawing program to add text to AI images, or asking it to caption things (does it do that well?)?

I agree it does sound like a good technique/process.

same here, pressing button is way easier than typing something creative in the comments so I leave comments only when I really feel it, like something is really cute or i laugh hard

prior to ai you would see a lot of poor photoshop/bad ms paint meme things that would make up for their crappy production values with wit and charm. there is very little of that with ai (at least on here) but why is that? it should be easier with the tools that let you doodle the composition no?