Hi again! Long time no see but as many of you need I was studying ComfyUI for the last 2 months regarding my Magnific "decoding" project which well, gave me a lot of useful info, I was able to do their tiling method system, learned a lot of things in Comfy, blablabla. Try ComfyUI in the case you haven't done yet.
Now... back to trainings.
I present you my rendition to one of my fav. artists and big influence in my art (and probably everyone) since I was a child. That's H.R. Giger .
What can I say about him? A total blast of wires, circular forms, surrealism, dark and biomechanical parts here and there that make his work, insanely good.
This time I trained from different sources but instead of focusing on his awesome palette, I changed it making it "rainbowish" (iridiscent sometimes) while trying to preserve the entire aesthetics which was... a challenge.
So... HOW TO USE IT ****************** - Trigger is g1h3r and class is style - You can use on any checkpoint but even not popular opinion, I'd recommend you to try first in the SDXL 1.0 official checkpoint by SAI (however works on any others). - CFG is tweaked so it works best due to image styles at 3.5 to 5. 3.5/4 should work. - Clip Skip 1 train. - Sampler is not an issue here. DPM 2SDE GPU should be ok. - Subtokens are biomechanical, alien with "modules" as painting/watercolor. These ones are being powered on purpose. - Weight recommended from 0.7 to 1. Sometimes 1 could too much depending which model is doing the inference. However... if you aim to just add a bit of influence try 0.4-0.55 since this is extremely extremely flexible (you will lose a lot of alien-wires-bio parts) but to make some custom combos, really good. THIS LORA IS GOOD FOR *************************** - Making creatures
- Creating weird buildings and organic forms - Circular forms and shapes (classic of him) - Biomechanical parts, complex and abstracts - Dark atmosphere punch - Cybernetic stuff and alien things alike - Drawings and paintings of characters and scenarios with Giger's influence (palette aside) - Some 3D renders when cinematic style is aim - Landscape at weight <1 around 0.6-0.7-0.8 (depending prompt)
Note: Some more examples of what styles you can go are here
That concludes all. By the way, if you are interested in the Stable Diffusion 1.5 version it's also available but via Patreon (to gift something to people supporting us there, kudos to them!). It has different datasets but the purpose is the same. Just different options to everyone. Thanks for the reading and enjoy!
