newest picklescan does not show anything... reading the pickle file also does not show any dangerous imports

@malikxseto Good! Talking in general, when I have a checkpoint and no instructions, there is a way to inquire what "words" it knows most?
For example, I have a checkpoint knowing some celebrieties and some other no. There is a way to inquire if a specific name is inside the training or not?

@settima_ai that is a problem you could take a gen made from the base model and then use the interrogation extension. But unlike LoRA there are no metadata with the used tokens...

You can use the vocab.json of the base model but this does not tell you NEW tokens....

You might find them with luck and enough interrogation but there is no easy way to determine the tokens afterwards

@malikxseto The funny part is: if you generate a batch of "portrait of xxx", where xxx is a celebrity, you will see a sequence of images resembling, plus or minus, that celebrity. But if xxx is a unknown name, SD will generate e sequence of similiar faces, according to the combination of text character.s For example: "portrait of blablabla" will "create" an immaginary character called "blablabla" with e specific face! So, in many cases is diffucult to decide if a celebrity is known by SD, or it is simply inventing it!
But a checkpoint is trained to connect images to words, right? ...And there is not a place where these words are stored?