Here is the home of the zoo for the liminal suburbs contest winner loras.
Prompt using "liminal", "suburban", "suburban liminal", and many variations. It's tagged with plain English unless specified with "tags" or "tags only", those are tagged with specific tags related to objects on the scene.
Use the model's quality tags and necessary detail tags to increase quality.
If they're SDXL based you can probably just swap them around all SDXL based models for fun results.
Suburban Liminal SD15 - 1024x1024 - Epoch 10 - 20,000 samples~ Civit
Suburban SDXL10 - 2048x2048 - Epoch 10 - 50,000 samples - Civit
Suburban SD3.5 M - 1024x1024 - Epoch 10 - 20,000 samples~ Civit
Suburban Flux1D - 1216x1216 - Epoch 10 - 50,000 samples - CIVIT
Suburban SimNoobV4-Vpred - pending
Suburban NoobSimV4-Vpred - pending
Suburban PonySimV4-Vpred - pending
Suburban SimSDXL V4 Epsilon - 1024x1024 - Epoch 10 ~ 50,000 samples - CIVIT
Tags only
I trained this one particularly because SimV4 is more stable than the others for architecture.
Suburban Liminal epiCRealism VXVI LAST FAME - Epoch 10 - 20,000 samples~ Civit
Tags only
I chose this because the quality is very high and the outcome is very high quality for uniformity with objects and humans.
Suburban SimPonyV4-Vpred - not released yet dead model
Suburban Illustrious - 1216x1216 - Epoch 10 ~ 50,000 samples - CIVIT
Redone v2 1024x1024 -> tags only, hand tagged
Suburban PonyV6 - 1216x1216 - Epoch 10 ~ 50,000 samples - CIVIT ~ Pending
Suburban is additions that were contest winners from Suburban Liminal, and they are absolutely choice.
Description
FAQ
Comments (5)
Step 1: organize the liminal suburbs contest.
Step 2: reuse the images for the LoRA.
Step 3: ?
Step 4: PROFIT
Jk, will try soon.
:D
I think you should retrain Illustrious version on booru tags, for now it's trained on full sentences and because of that there are some issues.
I can run it through. I let Civit handle this one, so I can just run it through again with tags instead of captions.
Captions are more responsive to any clip, but the unets themselves aka illustrious was trained almost exclusively on tags so it does a bad job with captions in my ways.
That is also the reason why NoobXL V-PRED took so much training to understand even some basic plain english without completely gutting an image.
Done. Hand tagged.