//=time() ?>
wow using different niko reference gives better images, so the closer the face the better the image because my AI was trained on anime faces yet it can still generate character similar to niko very cool
#OneshotGame
you can try my lastest text to anime AI now! keep in mind it can only generate anime faces and sometimes it ignore the text but you can conditioned it multiple ways with multiples texts or images or both it's fun
here is the link https://t.co/ygkadas7sM
my new AI model trained on 300k anime faces still only trained for 5 hours and the best part is it knows niko because the encoder clip knows niko
my AI now generate better anime image from text by adding mean(mean(dataset_clip_image_embbedings) + clip_text_embedding)
it shift the clip text embbeding from clip distribution to my dataset clip distribution and the second mean normalize it
it's so cool my AI is better than i expected wow this only trained with 99k images for 18 hours imagine if it was trained with bigger dataset and train longer