Create Side-By-Side 3D images and cross your eyes to enjoy the 3D effect.

3DSBSV1 will attempt to make an image that, when you cross your eyes properly, should give a true and believable 3D depth effect for the virtual center image that your eyes perceive. 3DSBSV1 is more of a beta LoRA than I wanted, but it works well enough to release it for feedback. You can (usually) add "focus dots" at the top of the image to help with centering your cross-eyed focus. The sample images are cherry-picked as this is still a work in progress.

I was inspired by seeing a post on Reddit that showed Flux is already capable of creating some rudimentary 3D side-by-side images. The base Flux capability for these kinds of SBS images isn't very baked so this is an attempt to make it more accurate and consistent. Training was done on a wide variety of aspect ratios so pretty much anything from square to ultra-wide will give you some interesting results.

This LoRA is highly experimental, not perfect, and I am already working on a second version.

Usage:
Weight of 1 and flux1-dev-fp8 was used for all testing,

Trigger the 3D format by adding the trigger word 3DSBSV1 and tailoring the effect by including one or more of the following phrases in your prompt:

3D side-by-side for crossed eyes
3D image for the left eye and the right eye
Focus dots at the top of the frame (adds dots to help focus your cross-eyes)
VR through the lens (gives the inset effect of looking into a VR headset)

You can try any of the following aspect ratios depending on the effect you want:
1024 x 1024 - good for tall thin subjects
1344 x 768 - sweet spot for most subjects
1536 x 640 - super wide for landscapes and scenery
1216 x 832
1216 x 768
1152 x 896

Known Issues:

it sometimes adds the words "Before" and "After" text at the top of the images, which must be coming from somewhere deep in Flux's training data since it was not among my training images.
The "focus dots" randomly appear even when not prompted.
Flux's legendary prompt adherence seems to be taking a minor hit, I think because the aspect ratio is so wide for 3D images.

Search AI Models & Addons

very few results