Please follow the guidelines below for best results!
I always loved Elites, especially Halo 3's design. But Stable Diffusion didn't share my enthusiasm. So I got to work, and hoo boy. I've never made a LoRA before in my life, and I certainly didn't expect it to take over 30 hours and 20 iterations of tweaking and fine tuning. But here we are. Now that this is done I can get back to what I love doing most - generating images.
After much pain and suffering, this LoRA is pretty good at generating decent looking Sangheili... IF you follow these instructions!
EasyFluff Checkpoint highly recommended for best results.
Clip Skip can be used at 1 or 2 for good outcomes. I prefer 2 and most preview images use it.
Weight works best at 1.
img2img recommended when your results are almost right, but I probably don't have to tell you this if you're here.
Prompting Notes:
Always use tags "sangheili, alien" to start a prompt.
For nude sangheili, use the tag "nude" with no other clothing tags.
To have sangheili wearing only an armor undersuit, use the tags "bodysuit, elitebodysuit"
To have armored sangheili, use the tags, "armor, [your preferred color of armor] armor, helmet"
Using the tag "bodysuit" alongside the armor tags will attempt to generate an undersuit underneath the armor (as opposed to the sangheili wearing armor pieces over bare flesh)
Works with several directional indicators such as "from behind, rear view, from side" in any state of dress.
If the eyes aren't visible enough, use "yellow eyes" OR "reptilian eyes"
The tag "open mouth" can make the sangheili open his mouth more. Results not guaranteed.
Recommended to always use the negative tag "tail." For rear angles, the negative tag "ass" is recommended unless that's what you want (how anatomically incorrect of you). For nude images, sometimes the negative tag "nipples" is necessary, and for laying down poses, sometimes you will need "3 legs."
Description
Trained on Fluffyrock but works best on EasyFluff. This LoRA was mainly made for my purposes, but in case anyone else was interested I thought I'd share it anyway. Hope at least someone enjoys. I'd be thrilled if you posted your results.
FAQ
Comments (5)
Wort wort wort!
Very nice.
doesnt seem to work. i dont know if its the file or the website but all i get is a mess of blue and green slashed of color on generation
Unfortunately I never generate on the site so I don’t know exactly how to help. I know for maximum compatibility you’re supposed to use the base SD model for training but the only way I was able to get acceptable results was to use Fluffyrock, so maybe the model you’re using isn’t compatible? Try using easyfluff if you’re not already just for testing’s sake. If you get good results with that, then my Lora’s compatibility just sucks. I wouldn’t be surprised if that were the case. I’ve been making images almost a year, but Loras for a week. So I’m a little out of my element with this 😅
To be a little fair to myself I have used many Loras that only work with certain models so this isn’t completely out of the ordinary, but it’s also fair to expect more from your Loras.
I really burnt myself out doing this so I need a break from Loras, but perhaps I’ll try revisiting it using base SD 1.5 as the training model sometime.
its true, i use Stable Diffusion GUI (by automatic) but whenever i try to use this lora, i only get a green figurine or nothing.
i tried using your promts or settings but i always got the same results, even if i used your recommended model
Very good
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.