Blue 24/7-- aka blue247.safetensors -- is a LORA designed to achieve blue skin without causing color leakage to the background, hair, or eyes of your character and without hijacking the style of your checkpoint.
To use the lora, use the following terms: blue_skin and (optional but it was trained using this term as the class prompt) 1person. (Blue_skin can be replaced with blue skin, though results may be degraded.) Blue skin too pale? Emphasize it as (blue_skin). Want a darker color, such as for Drow elves? Do (blue_skin) and either (dark blue skin) or (dark skin) depending on the checkpoint. I recommend using the lora at .8 str, but that can change depending on checkpoint.
Because I emphasized ensuring that ears would be blue, the LORA tends to favour creating elf ears. If you do not wish for elf ears, place (elf ears) in the negative prompts. :) I further recommend you place (blush), tattoos, and scars for the negative prompts on certain checkpoints if any of those begin presenting issues.
Note on if you copy my image prompts: I renamed the model from blueblend247 to blue247 but they are identical. Adjust the input name accordingly. :) There are also some resources that are not available on CIVITAI and I am sorry but they are not mine to re-release because I do not have the permissions to do so.
This LORA is a merger of three LORAs I trained on three checkpoints, including both NAI and SD 1.5. As a result, it would work with most any checkpoint merger. However, some custom checkpoints and some LORAs are very aggressively biased in their training data and will override blue skin with their own defaults. A workaround is to gen a starter image on a different checkpoint then img2img to your final goal.
Thank you to Dan and Jordan for testing the LORA, to Part_LoRAs for their colored skin lyco that inspired me, and to all the people that have worked on SD and the web interfaces, models, and guides we enjoy.
If you want to support me or offer feedback, suggestions, or requests, please drop a comment below!
An alternative to my LORA is the lyco coloredSkin-08 by Part_LoRAs that does a variety of skin colors, not just blue. I began work on my LORA prior to this updated release because the older version had some issues with color bleeding into other aspects of the image. I have not done any deep testing of this new lyco but it is worth mentioning as an alternative. To avoid cross-contamination of datasets, my LORA was NOT trained on images generated with any version of the coloredskin lyco so it is worth experimenting to see which works best for your particular needs.
Description
FAQ
Comments (9)
Ah man, for a second I thought someone made a Tau model haha
Haha, I tried genning one just not to test it but the checkpoints I use are very dedicated to genning people with noses. I would recommend using control net with an image of a tau and the line-art process!
@cootie yeah no prob, just thought it was funny because people have done almost all the 40k races but not Tau
Absolutely perfect, works flawlessly with multiple other lora
Any chance at Green/purple/red? ;)
Red is the next on my list! Interactions between red skin and negative embeddings mean I need to be hyper specific about the input images, however, because they like to correct red skin to over-exposed tan or brown skin instead, or leave great big splotches. If training goes well, I'll have it out in around a week. :) Thank you very much for the review, comment, and mage post! Love the picture!
Getting the inevitable "da ba dee da ba daa" comment out of the way.
and everyone: https://www.youtube.com/watch?v=zA52uNzx7Y4
It's true. I did start humming that.
thanks a lot