Posts

@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??

@Thoryn
Personally I hoped that more people would’ve let prompt remain attached to their uploads, but there are plenty of prompts to pull fro lm sites liek civitai

@Background Pony #D936
I’m not sure if such a thing exists, but you can ask Perplexity.ai about various terms:
https://www.perplexity.ai/

There is also this guide by MareStare (WIP):
https://twai.dev/welcome

does anyone know of a glossary of generative ai terminology? am stupide and not get what is being talked about

@tyto4tme4l
thanks. it is easy as you said

@Background Pony #BF52
You can try Bing Image Creator (powered by DALL-E 3). Super easy to use and can generate decent SFW ponies.
https://www.bing.com/images/create

what way do you recomend to genereten images to a complete newbi? I am trying for sfw scenes since am in school compueter

@Thoryn
As an added tip, there are ComfyUI nodes that can save your images without metadata in the first place. I use these pretty exclusively because I don’t want people confused by my workflows and wildcards.

@tyto4tme4l
@Lord Waite
Good tips, thanks!

@Thoryn
You can also view and remove metadata using this site:
https://jimpl.com/

@Thoryn
JPEGs could also have some metadata in them. If you’re particularly worried, you might download a tool like exiftool, which will lets you look at and modify the metadata yourself.
https://exiftool.org/

Any preference as to what file format should be uploaded?

PNG: Intact prompt metadata, but big file size which will impact storage, as well as bandwidth for both server and client.
JPG: Negligible quality loss due to compression, much smaller file size with regards to storage, faster to serve and download.. but prompts would have to be posted as a comment if they were to be shared.

Edit: Since I am paranoid that e.g. Windows account/user info data may be stored in the images, combined with noticing there’s a tag for prompt in description, and it will save on space and be more speedy, I will opt for jpg personally, unless told otherwise.

@Zaknel
no