I acknowledge the existence of other HR Giger-inspired models on civitai.com and express my admiration for their creators' work:
[1] Giger's Alien Factory by sebjclarke - A great checkpoint for generating xenomorph images.
[2] Biomechanicals (HR Giger) by grafikzeug - A LoRA model for biomechanical aesthetics that I'm still trying to tame.
[3] H.R. Giger Arh+ by ThatBonsaipanda - A 3MB LoRA trained on high-quality images.
[4] Hans «Rüdi» Giger style by Mince - An excellent trained checkpoint.
[5] xGiger by cobraeconomics and [6] hrgiger-drmacabre - Textual inversions I have yet to explore.
My appreciation goes to all contributing authors.
Regarding version v1.0, I just merged [3] and a LoRA extraction from [4], not realizing that [3] prohibited sharing merged models. Because I'm an idot. Consequently, I removed this version.
Over a weekend, I attempted to create a workaround replicating v1.0 by merging my trained model [GigTrained] and [Hans «Rüdi» Giger style 9MB LoRA] which is a 9MB LoRA extraction from [3]. I was fiddling with the weights between them too long. Then I just decided to share them separately, so the user will fiddle with the weights instead of me :D.
I recommend experimenting with these works, especially combining [Hans «Rüdi» Giger style 9MB LoRA] with [3] for satisfying results.
Also check [Giger-ish alien-like creatures (5MB LoRA)] -- A LoRA extraction from [1], creating xenomorph-like creatures (but no luck with the xenomorph itself).
I plan to continue refining my contribution in the future.
I offer all my content at no cost to you. Should you wish to show your appreciation and support, kindly consider buying me coffee via the following link: https://www.buymeacoffee.com/konyconiN. Your generosity is greatly appreciated!
Description
FAQ
Comments (15)
You are a genius with ideas
I would love to use one of those coffee machines each morning. :)
Bruh. I specifically labelled the LoRA as "please don't make merges of this because I spent several hours creating it". But it's all good. The first sentence in the description just flies out the window. :D
Oh dear, my apologies. It appears I completely overlooked that.
@konyconi Don't worry about it. o/
Fixed :]
@ThatBonsaipanda would you authorize/allow this fine gentleman to host v1 again? I feel like it was an awesome Lora and with permission could be an extra take of your version.
@zx96 No need. You can get the same results simply by taking the [Hans «Rüdi» Giger style 9MB LoRA] from here and ThatBonsaipanda's LoRA and use them both with the same weight. Alternatively, (if you can do it) merge them with weight 0.5 for both. That way you get exactly v1.0.
@zx96 Sure, I mean, it's no biggie when peeps make their own version of it. It's just really 'meh' feeling when you've worked hours upon hours on it. :D
I'm actually trying to learn how to make a proper LoRA and I have reserved about 250 high quality Giger shots for scanning. I'm hoping to make 512, 768 and 1024 versions but so far with the tests, the LoRAs don't "activate" when used, even with the trigger words and I don't know why.
@ThatBonsaipanda can you elaborate what is proper LoRA for you? I have used automatic1111 dreambooth LoRA option that is built-in, but now I get better results using lastben's dreambooth on colab and extracting LoRa afterwards using LORA extractor script from sashaok123 Github's page. Feels a lot stronger when training faces, havent tried styles though
@coudys My biggest issue was getting the learning rate and image repeats to properly converge at the end. The LoRA usually came out undercooked or just didn't activate. I'll get it working at some point and will post results. Been using a colab trainer and now recently local installation for the training.
prohibiting derivative work from your derivative work. lol. wut.
Keywords?
[GigTrained] does not require any keyword. [Hans «Rüdi» Giger style 9MB LoRA] has keyword "hrgiger".
Hey there, appreciate the shout-out! Your LoRA looks very good. Will give it a try soon!