Preview images inside kohya look way differently than images generate inside a1111, using the same checkpoint? What I'm doing wrong?

[dilectiogames]

I'm trying to train embeddings for SD1.5,

Here is how a test image automatically generated by kohya looks, (I've added the censorship so I can upload it safely)

and here how using the same prompt ( a single word the embedding name same one used in kohya Sample Prompts box) the same checkpoint, the same sampler inside automatic

what prompts kohya is using to get those images, there are hidden prompts? I really doubt this but I really don't understand the difference.

Second thing is that the embedding is really small in file size is 1.614 bytes, looks to me there may be a mistake when saving the embeding? Do you know where i can increase this size? Maybe I've forgot to increase it and it doesn't have enough data?

Did kohya internal generate uses the file saved on this or something cached somewhere?