I recommend using this LORA at a strength of 1 or more. In my testing (LORA strength 2) seemed to work the best. As well I recommend pairing this LORA with a 3d anime checkpoint.As well it should be noted that the model seems to work poorly on simplified checkpoints (2d anime) and models geared for realism. (Maybe you can get it working IDK)Trigger word: tentacle_helperAlso of note: tentacles1, tentacles2, tentacles3, tentacles4, tentacles5 Above are all variations that have (unfortunately) little effect.If things get too blue negative prompt: blue theme, blue background. But that shouldn't be necessary.The idea was to use real life tentacles to make a LORA. I used pictures of sea anemone &amp; starfish as subjects. I gathered the training pictures from an aquarium in Alaska (Shoutout Alaska SeaLife Center).While I am happy with the results, I failed to create the variety I was looking for. Instead what you get is tentacles that look like the underside of a sea star. I suspect that my training data (that was) skewed toward the underside of sea stars is the cultrate. Perhaps one day I will make more tentacle LORA for other kinds of tentacles.

Tentacle Helper

v1.0 Over trained (100 steps per image)v1.1 Under trained (20 steps per image)v1.2 Trigger words don't seem to be working, restructuring training prompts and adding a trigger word.v2.0 Working, not as much variety as I would like. I assume tentacles1 is unique though perhaps not? Good enough for a release though.

Files with SHA256: 283bb9d3893739bcb6c1411d4b7d21382b0f4bae692e19c996a8db9db4547e48