Spaces:
Running
Empty string input result
I've noticed using an empty string as the input usually results in images of a woman wearing a sari dress. Does anyone have an explanation for this result? Is it just the most common training image maybe?
Edit: correct dress name
This a majority non-deterministic system. You can't get the same result. Ever. as far as I understand.
And yes, maybe there are several million pairs of training images in the model with no caption.
But who really knows :) ?
Oh yeah, I don't mean the exact same image. But the same general theme. Just seems to be its go-to scene to draw with no input, beside random textures.
Every image correspond to a point in a 500-dimensional space. I think that woman in sari dress is just what happens to be represented in the point near the default values.
Ahh! images without captions...that does make sense.
BTW, here's a link to a newer discussion where I requested users to post their "blank" runs.
https://huggingface.co./spaces/dalle-mini/dalle-mini/discussions/619