change training credit logic: 1) when starting training new model, remove 1 training credit; 2) if fails, add back 1 credit, 3) on training retry remove 1 training credit #photoai
count currently training models to avoid race condition where user has 1 training credit, then trains 1 model, but before the model is finished, trains another one, and then only when they finish they decrement the training credits #photoai
try retrain SD1.5 model with Flux with the same training ZIP (with stuff like face crop, body crop, stuff that worked for SD1.5) and discover that retraining old SD1.5 models with that training zip won't work well so maybe we should not #photoai