I think it's overtrained on forward facing close ups.
But the likeness seems to come through quite well.
Was thinking of doing a Slipknot mask LoRA at some point. Just throwing a bunch of untagged images and seeing if we get a mask generator?
Description
PICS: 19
BLIP: 5 - 75
min_snr_gamma: 5
steps: 10% 34/340 [00:46<07:00, 1.37s/it, loss=0.0779]
steps: 20% 68/340 [01:42<06:51, 1.51s/it, loss=0.0874]
steps: 30% 102/340 [02:39<06:11, 1.56s/it, loss=0.0818]
steps: 40% 136/340 [03:34<05:22, 1.58s/it, loss=0.067]
steps: 50% 170/340 [04:30<04:30, 1.59s/it, loss=0.0838]
steps: 60% 204/340 [05:25<03:37, 1.60s/it, loss=0.0828]
steps: 70% 238/340 [06:21<02:43, 1.60s/it, loss=0.0835]
steps: 80% 272/340 [07:17<01:49, 1.61s/it, loss=0.0756]
steps: 90% 306/340 [08:13<00:54, 1.61s/it, loss=0.0698]
steps: 100% 340/340 [09:19<00:00, 1.65s/it, loss=0.0684]