Big Liminal

Big Liminal V1 is the result of a multitude of gathered liminal space packs, generated images from additional AI, people's personal artwork, and an accumulated dataset based on findings.

https://huggingface.co/AbstractPhil/Liminal-Full/tree/main

All Big Liminal weights hosted on HUGGINGFACE as they come in and ported to Civit when I have time.

All images used are at least size 1024x1024, at maximum 4096x4096.

Target size at least 1024x1024 loras with the goal sizes around 2048x2048 for some models that can support it, and smaller for those that cannot support it without a lot of training.

All of them trained at MIN_SNR 7 BUCKETING at the exact same learn rates.

V1:

Big Liminal 976 liminal images trained at repeat 5

Suburban 111 suburban images trained at repeat 7

Big Liminal SD15 - pending

Big Liminal SDXL10 - 1216x1216 - Epoch 10 - 50,000 samples - Runpod

Big Liminal SD3.5 - 1024x1024 - Epoch 6 - 30,000 samples - Runpod

Big Liminal SimSDXL - pending

Big Liminal NoobSimV4-Vpred - pending

Big Liminal SimNoobV4-Vpred - 2048x2048 - Epoch 13 - 48,000 samples - Runpod

Big Liminal PonySimV4-Vpred - not released yet

Big Liminal SimPonyV4-Vpred - not released yet

Big Liminal Illustrious v10 - 2048x2048 - Epoch 5 - 50,000 samples - Runpod

Big Liminal PonyV6 - 2048x2048 - pending

Big Liminal NoobXL-VpredV10 - 2048x2048 - Epoch 5 ~ 25,000 samples - Runpod