i got great results but they are a bit too spicy to post :D thank you for this, i will try some more stuff. The colab strangely didnt't work well but the extension in automatic. The filewords option didn't work, i created txt files along the images but it looks like it didn't read them like for embedding. Does it only work when the words are in the file name itself? i then used a prompt with 3000 steps, 200 for each image on 20 images, i reduced the default learning rate slightly. Took only about 1/2 an hour on 3080 rtx 8 gb vram.