crashbandit said:
Okay. I haven't thought about going to basics that much. I've been focused on preserving the whole image, and using AI to perform the corrections or avoiding the error in the first place. I focus a lot on movement and the scene so keeping everything in is generally important to me. I probably need to rethink this.

well, if you don't want to change the image and the image doesn't meet e6ai's standards then something's got to give and it's probably going to be you. don't think you have much option other than not uploading the images in that case (or i guess letting them get removed, but would be a jerk move to create extra work for the janitors by posting stuff that they likely have to reject).

My impression is the better quality generations we're seeing today are due to a modification/add-on that can tag specific portions of the image, and using a base such as a sketch. Not sure how true that really is. The really good stuff seems to be from custom models and algorithms, but I don't have a lot of insight into those.

i haven't heard anything like that or used that sort of thing personally. models in generally have evolved, so fairly recent model is naturally going to be better at stuff like not creating abominations/messing up anatomy than a model from a year ago.

I'm used to the idea of going in manually and redrawing or manipulating an area to fix issues, which is super time consuming and I'm not skilled at. How do people go in and quickly make changes to say a hand or the eyes? I've seen some folks turn things around in a couple of minutes, which very likely moves it beyond doing it manually.

they're almost certainly using inpainting - you can mask off a specific region like a hand or an eye and just regenerate that part without changing the rest of the images.

Novelai seems to be a year or so behind from what I can tell. I like novelai as it seems to generate better overall anatomy and texture, but it really seems to not do eyes well. It also doesn't seem to handle fingers and toes too well. Still trying to get something working on the desktop so I can try out some of the other things that exists today to see how they really compare. A lot of what I'm seeing generated I'm guessing is trained on a narrower dataset, and it produces the same errors I see a lot of artists produce. In fact that's getting worse, probably because it's more prolific.

the main thing for local generation is having a good enough GPU. also for stuff like inpainting you don't _necessarily_ have to use the same model as the original generation, just something close enough that the changes will blend in. so if you really like how novelai does the initial generations and you can't find something else that works for local generation you could still do your cleanup locally with different models potentially.

Also before the original question is lost. Is there a way to view the image along with the deleted post so I can identify what was actually deleted? This is something I really want to figure out.

not as far as i know, and i agree it's pretty annoying that you can't. you basically have to figure it out from the image tags. it is possible to search for the images that have been deleted directorname status:deleted and approval is super timely these days so you shouldn't need to remember image tags for more than a day or so. also there probably isn't anything stopping you from putting some kind of hint to yourself in the description that will help you identify images if they get deleted.

it would be better from a user perspective if the user could see the deleted image even if no one else could but something like that would require changes to the site code which is probably not very likely. there are also reasons why it doesn't work that way. extreme example: if someone uploaded something like child porn, we would not want to keep serving images of that from the site to them