Attempted strange expierement of adding tokens to the captions, to see if it might improve things.
PICS: 47
TOKEN: ugly, brunette, blurry, jpeg_artifacts, out_of_focus, low_resolution, low_quality

BLIP: 30 - 100

steps: 10% 79/790 [01:50<16:38, 1.40s/it, loss=0.129]

steps: 20% 158/790 [03:49<15:16, 1.45s/it, loss=0.124]

steps: 30% 237/790 [05:46<13:28, 1.46s/it, loss=0.142]

steps: 40% 316/790 [07:43<11:35, 1.47s/it, loss=0.136]

steps: 60% 474/790 [11:38<07:45, 1.47s/it, loss=0.13]

steps: 60% 474/790 [11:38<07:45, 1.47s/it, loss=0.13]

steps: 70% 553/790 [13:36<05:49, 1.48s/it, loss=0.142]

steps: 80% 632/790 [15:33<03:53, 1.48s/it, loss=0.123]

steps: 90% 711/790 [17:31<01:56, 1.48s/it, loss=0.144]

steps: 100% 790/790 [19:28<00:00, 1.48s/it, loss=0.134]