This lora create SBS 3D images that can be viewed by the crosseyed method
The phrases: "Stereoscopic SBS side-by-side 3D," At the start of the prompt helps a lot.
Check image prompts as examples.
V2 Update: V2 is leaps ahead of V1. I refined the dataset by getting rid of images with minimal effect and parallel view images that slipped through the first time. I also tested the produced loras a lot and came to the conclusion that 2400 steps lora is the most ballanced of all. For some reason, even after 3600 steps Krea 2 refuses to increase the depth of the effect. So these are the best settings I propose in order to have the best results:
Lora strength 1: minimal 3d effect, and minimal NSFW in SFW prompt.
Lora strength 1.3-1.7: Very good 3d effect and minimal NSFW on SFW prompt.
Lora strength 1.8-2: Best 3d effect and more NSFW (nudity) images on SFW prompt. Possible appearence of text in corners depending on the checkpoint.
The level of NSFW on SFW prompt depends on the model. Heavy NSFW models tend to undress subjects more. Censored models keep things clean.
In NSFW prompts you can use Lora strength 1.3-1.7 with excellent results.
Description
FAQ
Comments (9)
I don't see any parallax shift in your example images.
Check out some of my examples if you're not sure whether it works. I've been running these at lora strength 2 to push the effect harder.
bennyboy_77 is correct in his comment about KREA 2 having its own training on SBS. If you test the same prompt/seed with and without the lora you will see that Krea 2 also creates the image but (at least to my tests) the parallax is even less than the one produced by this lora. I plan to refine my dataset to images with even stronger effect and retrain for a version 2 and see how it comes out.
In my limited testing so far, I'm getting a small amount of genuine 3D effect but not as strong as other loras I've used before. I've tried pushing the lora strength up to 2 and got a few decent-ish images but it tends to double up the subject in the image as well if the strength is set that high.
Also, although the default format for this lora is in the cross-eyed SBS format, sometimes elements of the image seem to be in parallel pair format so, for example, a wall behind someone might look closer than the person. I wonder if Krea 2 itself has some limited SBS knowledge and is applying it on top of the lora training to create a weird hybrid effect?
Perhaps it is worth retraining another version of the lora in the parallel pair format (which I think is considered a default for 3D viewing). This might sit more happily with any 3D SBS knowledge that Krea 2 already learned. (I'm just guessing here btw!)
I'll post a few of the images that turned out better because the examples provided by the lora creator seem to be lacking any 3D depth as far as I can tell (maybe my eyes are just getting old!!).
Sorry - forgot to thank the poster for his/her lora then left a long message that sounded kind of negative! If I'm nitpicking, it's because there's a lot of promise with this lora and I'll definitely be doing some more experiments. Thank you OP! Example images seem to have appeared now and I'm quite happy with them. I think all were done at strength 2 for the lora and prompts emphasized near and far objects to push the likelihood of getting strong 3d effects.
Thank you for your input. From my tests Krea 2 seems indeed to have some sbs training, but all the images that came out without the lora had even less 3d effect. Previous training in FLUX 1D in an unpublished lora gave better results. I guess I need to refine the data set and retrain and see what happens.
SBS is something I've tested extensively, the best way to do it is probably not with a LORA, it requires really precise alignment to get the effect just right. An app I've used before that produces the highest quality and top-tier results for 3d SBS and also works with video and everything you can throw at it is called Owl3D, unfortunately you do have to pay for it, but in my experience I haven't used anything else that comes close.
@Mohzaic Owl3d does excellent with images, but Ive not gotten a lot of depth with video. Not sure if its me. That is with my 5070ti on top of that. Generated images from Krea2 then used with Owl3D app afterward totally look so amazing.
@brianr1973871 Might need to tweak the settings within the app a bit. I had really successful results with video.