Pony version released.
This Lora generates stereograms that can be viewed stereoscopically with cross-eyed viewing.
v2 released : Higher success rate than v1, more stable picture including background (I believe).
Trigger word: stereogram
Some Interesting keywords: pointing at viewer, reaching towards viewer, aiming at viewer
You may consider including keywords such as closed mouth, open mouth, parted lips, etc. to align the mouth expressions in the left and right images
More detailed prompt seems to stabilize the picture..
Suitable weights: 0.5 ~; seem to vary from model to model.
Try to find the best value starting with a small value.
It appears that higher weights do not necessarily result in more stable output.
Recommended resolutions: 768x512, 768x768, 768x384
Inpaint can be helpful in correcting minor mismatches between images (e.g., mouth expressions).
The yield is not very high. It is even more difficult with character Loras.
It may be better to use LoRA Block Weight and set other Lora to OUTALL or OUTD.
Please share your creative work!
Description
Improved stereoscopic output than v1, with more stable backgrounds.
Suitable weights seem to vary from model to model; even 0.2 ~ can be effective in some cases.
Larger weights will affect the style of the image (more realistic). Try to find the best value starting with a small value.
e.g meinahentai: 0.6 - 0.7
Clothing tends to have a fantasy look and may need to be corrected at the prompt.
Inpaint can be helpful in correcting minor mismatches between images (e.g., mouth expressions).
FAQ
Comments (5)
Yeah, definitely difficult to pull "sound" yields, but very cool concept. This might operate better as a Lycoris?
I think this concept can be experimented with a bit more, and Lycoris might be an option to consider.
I love the concept but I've never got the crosseyed method to work. Perhaps a red/blue version?
My priority right now is to stabilize Lora for the cross-eye method.
Anaglyphs (red and blue glasses) may be an option after that. Another option is parallel method stereograms.
By the way, it seems that CivitAI already has an anaglyph Lora, but I haven't tried it because I don't have the glasses (and it seems the author of Lora doesn't either).
https://civitai.com/models/91144/tangbohu-stereoscopic-red-blue-simulator
So you know how when you go cross-eyed you see double, right? You're looking at a double picture and trying to make the "double-doubles" line up so the 2 middle doubles overlap creating 3 pictures... one in the center and two in your peripheral vision. You're really trying to see "3 pictures" by the end, with the center picture being a composite of the two inner doubles created by crossing your eyes. You can try that method or a shortcut which is more hit-or-miss which is to put the image on your monitor, then bring up your phone and read something 6 inches away from your eyes. Then without changing your focal plane, lower your phone. It should at least get you close. Also, there's an extension that uses depth maps that more or less does this automatically and it can do either cross eye method or vr headset side-by-side or top-bottom 3d. You can download it from within Automatic1111 or get it at https://github.com/thygate/stable-diffusion-webui-depthmap-script and drag&drop it into extensions.
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.