The dataset was annotated with descriptions starting with "Photo of Max Verstappen, a man ...". In the standard Flux dev model, the token "Max Verstappen" tends to generate an image of a man wearing a racing suit, regardless of the variations in clothing I carefully selected in dataset. To avoid this issue, you can either exclude "Max Verstappen" from the prompt or explicitly describe the clothing you want him to wear.
When I get BUZZ again, I plan to retrain using a unique token to fix it.
The dataset was annotated with descriptions starting with "Photo of Max Verstappen, a man ...". In the standard Flux dev model, the token "Max Verstappen" tends to generate an image of a man wearing a racing suit, regardless of the variations in clothing I carefully selected in dataset. To avoid this issue, you can either exclude "Max Verstappen" from the prompt or explicitly describe the clothing you want him to wear.
When I get BUZZ again, I plan to retrain using a unique token to fix it.