Krea2 - v1.0:
(There is a simple example workflow you can download under "optional files")
This Lora will create side by side images creating the illusion of stereoscopic vision by creating left and right images. This version is for text to image creation.
The database consist of self shot sbs 3d images with my KanDao 3D Kamera. That is also the reason it will deliver rather dim lit and very realistic images. You can go down to Strenght 0.9 to get 2D images to work properly.
Using Krea2 zriggers are important:
SBS_Stereo_img. stereogram. On the left: image1 (keep the image1 exactly the same). On the right: slightly shifted second image with slightly rotated perspective.
Description
t2i only
FAQ
Comments (13)
Looks cool, I have done SBS 3D a lot before with diffusion models but never using a lora. I will test it out now.
thanks for posting. the result tells me xou used a 2 pass workflow? I'll try to provide a workflow asap Looking forward to new posts then ;-)
Here you go. Fresh workflow 4u. I'd appreciate it if you could replace the posts you made with those you make with the workfllow :-)
@Adaptalab0r No, its just a quick single pass output with my model, and your lora at a lower weight than 1. I will test out your workflow and compare.
@GlowingGuardianGirl No, there isn't. and depending on how it is done I would rather use LTXV2 for something like that. I'm tempted to make one for LTXV2 but the sbs material I have has lots of watermarks. I don't have software to remove them fast and easy. So I probably won't do it sorry.
@GlowingGuardianGirl well. my technique is straight forward but not very good. i use side by side video capture them and hope for the best. i wrote to the author of the best stereo image lora there is, but he didn't answer to explain to me how he did it, but i made a wokflow using his Lora to conveniently convert frame vy frame into Video. you can check it out here (qwen version):
https://civitai.red/models/2720656/stereoscopic-sbs-3d-image-workflow-flux2-klein-9b
Apologies if I'm not being very bright. Is this to be used as purely text to image (which could certainly be amazing if it works well) or is there a workflow that allows image to image stereo conversion? Also, all these examples seem to be in cross-eyed format rather than the SBS format normally used for VR headsets etc. Thanks in advance for any info you have.
well, it's txt 2 image only. I'll try to fix a simple workflow soon. usually sbs 3d viewers allow to switch left and right. maybe i can do something about it. the WF I used is huge, because I use a all in one WF.
I took the official template and transformed it into a simple example workflow you can download under "optional files". It saves the image as crossview AND 3DTV left-right swapped image. Enjoy ;-)
Amazing. Thanks for doing that. I'll have another go with it.
@Adaptalab0r I've been using this on and off for the past week and getting some good results. As always, it's a case of rolling the dice a few times to get a good outcome. I'm quite into the fact that it also adds an "amateur photo" vibe to any generated image.
I wonder if there's a way to combine the effects of a lora like this with a depth map to more precisely control the shape of the environment. I tend to find that the person/object that is the central focus of the image gets good results but there are oddities in the depth towards the edge of the image (which slightly breaks the stereoscopic illusion).
Thanks again and I look forward to any future stereoscopic loras/workflows you come up with.
@bennyboy_77 Thanks fpr the feedback. I like the amateur realistic style and take it as an happy little accident, as the Qcam Camera is rather dim lit and grainy. I have the exact same issues with the peripheral image setting and hope next versions will be better, but they'll have to wait. I think you might get a notification if there is a new release.
@Adaptalab0r Thanks. Yes - I've ticked the little bell icon to get notified of any future versions. I've added a couple of images using a new lora I just made - Lara Croft promo model/cosplay - early 2000s style. I might actually post that lora to Civitai as it turned out pretty well.