SDXL version of the H.R. Giger LoRA I released earlier. I let this run for 6000 steps but as it is such a unique style, I might need to let it run 6000 more at some point, there are some stuff that it sometimes renders that is a bit janky, detail wise. I'm also going to try and run one LoRA that has captions, this one is just pure style with no captions so it goes a bit wild sometimes.
Made from a curated stack of very high quality samples of Giger's work, some were photographs of the actual artworks, some were scans of large prints.
Description
First test run of SDXL training, No captions, 6000 steps, Prodigy with Cosine Annealing LR
FAQ
Comments (4)
This LoRA is awesome! Nice work! I've combined it with another LoRA trained on myself and it's so fun to watch myself get mutilated in a Giger biomechanical landscape. Works pretty seamlessly with other SDXL LoRAs, as well.
What checkpoint do you recommend??
You did not put the checkpoint used or the Strength of the Lora
Yeah, the downside of using ComfyUI is that it didn't bring in the details. The LoRA should work with almost any SDXL model and changing the strength varies the images as you would expect - 1.0 should not overburn, but apply the aesthetics in 'full'. Most of the examples are img2img examples, but if you take the original SDXL model with a simple prompt and set the LoRA strength to 1.0, it should produce Gigeresque monsters.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.