- Jan 24, 2020
- 1,401
- 3,793
You can always use the Interrogate functions in A1111 img2img, it's much slower ofc because you have to do one image at a time. When I start on a fresh prompt, sometimes I use an image with both clip and DeepBooru and then take the best from both, then I adjust and add my own stuff etc. Even though it's a big pain in da ass it's well worth taking the time to get really good captions when training a Lora, because good captions is very important to get a good result. According to what I have read, the training is more sensitive and less forgiving to bad captions than the quality of the images used for the training. Also if you are not careful a tag that is present often enough because of an item in the images that you are describing is consistent in many images, this can create unintentional trigger Words. I wasn't aware when I trained my Kendra Lora so "headband" became one of the trigger words.Don't have time right atm, but will try later. As quick run down:
- <tbd>
- horrible, it was just blip without beams as that is still broken for me, yay
- settings is exactly the same as i used in the lora i posted a while ago, constant, 0.0001 lr, 128/64 network, adamw 8bit, cyber_3.3, can't remember there being much else settings atm
- prompt for generation was basically just "woman dressed for jungle exploration", AI clearly think you don't need too much clothing, probably due to heat and you see she looks drenched in sweat
- "nothing", coloring the AI picked up itself, it really likes using it for clothing, rest is more then likely to not having captioned anything about the clothing. captions are so basic they mentioned next to nothing besides, woman, <optional pose>, <optional location>. Lack of beams with blip is horrible.
In some images there was a fence in the background and it became trained into the Lora, so sometimes I need to add it to negative prompt with a strong weight to avoid it in the generated images. In order to get good sample result in the training you have to describe what is in the image and there are only so many ways you can word it, so then it can't be helped. You just need to do your best and cross your fingers. Sometimes..