I should have found more diverse facial expression data. In prompts use something like: <code>small adult man</code>.

Peter Dinklage

PICS: 24BLIP: 5 - 75<code>min_snr_gamma</code>: 5steps: 10% 42/420 [01:00&lt;09:07, 1.45s/it, loss=0.102]steps: 20% 84/420 [02:08&lt;08:35, 1.53s/it, loss=0.0985]steps: 30% 126/420 [03:16&lt;07:39, 1.56s/it, loss=0.0947]steps: 40% 168/420 [04:24&lt;06:37, 1.58s/it, loss=0.0937]steps: 50% 210/420 [05:32&lt;05:32, 1.59s/it, loss=0.0959]steps: 60% 252/420 [06:40&lt;04:27, 1.59s/it, loss=0.0979]steps: 70% 294/420 [07:48&lt;03:20, 1.59s/it, loss=0.088]steps: 80% 336/420 [08:56&lt;02:14, 1.60s/it, loss=0.0946]steps: 90% 378/420 [10:04&lt;01:07, 1.60s/it, loss=0.0954]steps: 100% 420/420 [11:23&lt;00:00, 1.63s/it, loss=0.095]

Files with SHA256: cbd528e3cc7d978fa57dfe0f1af36434c966b7361fb0fa2e67717ed2a02771d7