The training was conducted on 50 different pictures generated by sora.
checkpoint-e36_s396.safetensors