From 500,000 AI-generated illustrations, 24,036 were selected and trained for 4 epochs using LoRA with the LION optimizer at a learning rate of 3e-5. They were then synthesized with a 0.7 application rate. Afterwards, 515 illustrations, with modifications such as finger corrections, underwent an additional 200 epochs of training using the Prodigy optimizer.
From 500,000 AI-generated illustrations, 24,036 were selected and trained for 4 epochs using LoRA with the LION optimizer at a learning rate of 3e-5. They were then synthesized with a 0.7 application rate. Afterwards, 515 illustrations, with modifications such as finger corrections, underwent an additional 200 epochs of training using the Prodigy optimizer.