Liked by 22 users: afcybrad, Atraidis, Chicklets09, chrisrednsa, CN1888, commodore, Daiwai01, Dick Sneider, Glanzick, grosmeister, Hell_Knocker, java_head, Kluftinger, Loco696, lonk, martins, METALFUT, mocone03, Sporkyy, stacydonovan3, staticman, wallad
Originally Posted by spacebag
Wait, wait... This pictures are AI generated????
Depends how you want to define "AI".
This is a fancy "face swap" using AI (stable diffusion + safetensors checkpoint trained on a face + existing similar content).
It is a lot more complex than basic inpainting or copy/paste/retouch, but the entire image is certainly not created from scratch using "AI".
Most of the AI in this section is based on existing AI models from Civitai and LoRAs (or similar methods)... those are great too, and the results can be anything you type into the prompt. But they are very different and very artificial regardless of how great the base model used is.
Idek, those are neat, I have tried that too and was not impressed with the amount of garbage it generates.
I went a different route, there is plenty of existing content for almost scene you'd like, and that is what I have experimented with here.
Liked by 3 users: fulviobors,john_siris, preacher620
Oh, I understand now! So this are like good old fakes but AI assisted. So instead of trying to find the exact face that fits the position and lightning of the body donor, you train the IA with many pictures of that face and then you make a deepfake face swap. Just like they did with Luke Skywalker in The Book of Boba Fett or with Harrison Ford in the 1945 sequence of Indiana Jones and the Dial of Destiny. But instead of a sequence of video, you apply the technique in a set of pictures.
In my misunderstanding I was astonished with how real it looked to be completely AI and how good was the temporal consistency of all details in the body and the environment. This is something current image generation AIs can't do correctly (to be consistent in a set of images) but I think in a couple of years we will be able to generate a set like this completely drawn by the AI.
In any case, it is a great work!
Yea, "AI" is a ambiguous term... most of the posts in this section are using an existing AI model (probably downloaded from civitai) and an existing Person to combine it with (also probably an existing LoRA or similar format they also downloaded)... the creative part comes from knowing how to think of unique text or reference images to prompt it with that with that gets good results to create some kind of interesting new images.
Learning how to do all that well is a skill and requires some good hardware and/or ability to understand how to run the tools. I am impressed by some of the results people are posting here, but I also know how it works and isn't that hard unless they are actually training a new model from scratch and making new/unique embeddings of some kind to combine them with.
I also know a few people here are doing this on hard-mode and they deserve a ton of respect.
That said, is neat and amazing, but the results not realistic and are generally pretty boring to me personally.
You can clearly see the difference... like you said "astonished", my posts do not look possible by AI because they aren't using the common techniques used in posts in this section... I think the most interesting results are possible if we take a step back from most technical AI methods and do some "simple" faceswaps on existing content.
I chose to focus on generating content that contains niche but well-known faces in a realistic way + real content/scene. IMO, there is no need to create a full AI model and embeddings that you can prompt for any scene you want only to have look like a dumb cartoon in the end... there are already billions of images of anything I could imagine, so why bother with the prompting?
(maybe I am not as creative as others, but based on what I see here, the imagination of everyone here else is pretty basic and lots of people seem to like like that).
To be fair, what I am doing is even easier/faster than what is mostly being posted here. It is a different technique though and has it's own complications and learning curve.
What I do is very much "AI" also, but a different workflow with different tools and steps. There is no way to get these results doing old-school copy/paste in photoshop or whatever without a ton of work, AI tools do the heavy lifting and make it possible. It takes a lot of work to create this stuff, leaning more on art than compute power and AI prompting skills ... I have experimented both and know the differences.
Anyway, glad you found my posts interesting. You can find some more examples if you search my posts, most of them are not in this board anymore.
I will probably get back to posting in this board when I think I have something more interesting to show than the cartoons that fit in this board
those images are gone, can anybody reup pls?
Does anybody have more content like that?
I've only found a Maya P+Emma Myers AI edit. It's not too spicy, but it looks cool.
Liked by 1 user: Sporkyy
Posting Permissions
Reply With Quote