This is my Orc Girl LoRA, made at the request of a commenter on my previous LoRA. If you guys have any requests, let me know, and I'll see if I can do them.
This LoRA works with relatively little fuss at 0.6 to 0.8 weight. It has a hard time generating armor, but I suspect less weight to do the trick. I use this at 0.75 weight.
The keywords are 'orc', 'orc girl', 'colored skin', and 'green skin'. You can choose to have tusk or fang tags if that is to your liking. All images generated on my custom blends that are AOM3 based. All on the OrangeMix VAE and Hires Fix.
Description
FAQ
Comments (21)
Can you point me in the direction of the 4x_fatal_Anime_500000_G upscaler you used?
did you ever tag the skin color green in any of your training captions ? that'd possibly make the training ignore it.
I pruned the 'green skin' tag from my tags for a good reason. If you do that, then the specific concept I'm going for becomes implicit. This means that more of your model can handle the generation rather than the LoRA. This makes generation more malleable and less baked.
Having LORA's that require you to jump through 20 prompting loops is so annoying.. I love this.
@schneed yeah that's what I meant (that you shouldn't mention the green skin) perhaps I wasn't clear ;)
Could you do a Lora or checkpoint in the style of the artist “inputwo”? He draws really nice stylized butts and legs with exaggerated proportions.
The most I find of this artist is 41 images. The lowest amount of images I've trained on is 50, and that one never really turned out well. I'd like to do it, but chances are you wouldn't get a good LoRA. I do like the idea of exaggerated lower proportions though.
@schneed https://vk.com/public149863943 You can find it on the Russian-language website, it's full of his art. Good luck with the creation of Lora, it will be interesting to see the result
I have a collection of at least 150+ images of his, let me know if you're interested
I really like the background art style in this..
Do you think this Orc Lora could be used for just backgrounds or landscapes?
If not would there be some way to do a lora for landscapes / cityscapes in the art style of this Orc Lora?
I could do a D&D style background LoRA, but I doubt it would prove particularly effective on the models I use. I'll look into it.
@schneed Awesome, huge fan of your work. Thanks!
@DreamExplorer You could also prompt it with "people" negatives ( 1 girl, 1boy, people...)
To get the background, and use that prompt with the "AND" operator
@mandraw Did you try it? Does it work?
I think that depends on the dataset?
I will try this later today when I have a time for sure and report back.
@DreamExplorer I forgot to say, you'd need composable lora extension, but yeah just tested it works
@mandraw Thanks for your help.. do you have an example prompt / example? I installed everything but it's not clear from the README what I'm doing..
@DreamExplorer you should have a "composable lora" somewhere in your txt2img tab, usually after controlnet if you have it. there you enable it, then the prompt could be something like this:
1girl, red eyes, yellow skin AND forest background, detailed <lora:orcGirlLora_v10:1>
The "AND" should be all caps
@mandraw I get it.. this makes the LORA only apply to the part of the prompt that comes after "AND" so you can mix and match LORA's but only have them influence the part of the prompt you want...
I'm dumb.. but I get it now!
Fine lora, but it's still a pain to make tusks noticeable without actually drawing them and img2img.
a goblin girl lora would be also <3
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.