Introducing Nova Skyrift!
She's goth, she's fantasy, she's cyber.
Nova Skyrift is a very versatile custom LoRA trained to fuse crisp anime linework with neon-fueled cyberpunk vibes and photorealistic polish. Whether you’re crafting solo portraits, gritty back-alleys, or high-fashion editorials, this LoRA brings the best of anime, cyber, and realism into one package.
V1 TRAINED ON SDXL
Nova Skyrift was trained with SDXL on a series of 99 images chosen out of 500+ generated using Midjourney. Dataset is available for download.
The whole process was a learning project for me. I created a batch of images using Midjourney about 6 months ago before I started learning Stable Diffusion and ComfyUI - I had no real plan for them at the time. I got it down to 99 for the LoRA. After training Nova Skyrift V1, I did extensive testing. I coded - with the help of ChatGPT - a couple random prompt generators based around core scenes and themes. Here are some sample prompts from the HTML code:
Nova Skyrift Dual Toggle Prompt Builder
(Cyber Mode Toggle ON / Gothic Mode Toggle ON)
"1girl, solo, ancient crypt wreathed in incense and ivy, 1girl spinning beneath a red HUD projection in sheer bodysuit, burning candles beside analog screens, cyber-goth cinematic blend, lipstick smear and chipped nail polish, bare shoulders glowing in firelight, ultra-detailed, photorealistic, cinematic lighting, 4K render"
Nova Skyrift x Retro Sci-Fi Prompt Builder
(Nova Skyrift V1 combined with my Retro Sci-Fi Control Room_Illustrious_V3_NEW)
"1girl, solo, retro-style mission control beneath a volcano, 1girl in a chromed bikini armor with neural uplink visor, misty atmosphere with hazard labels and blinking lights, 1980s sci-fi pulp cover style with vibrant grain, badge-insignia across chestplate glinting under lights, robotic hand gently brushing her hair aside, retro sci-fi control room, ultra-detailed, cinematic lighting, 4K render"
I ran generations on 28 checkpoints, using a variety of LoRAs and prompts. I finished with over 600 images and realized it was going to take a lot to get this sorted. Again I used - ChatGPT - this time to help me code python scripts to automate the sorting. After a number of attempts, I ended up with a script that would sort the images by Base Model and LoRAs used, extract the generation data in a text file, rename all the images with the same checkpoint and LoRAS as "Gen" groups (Gen 1 - Gen146), then box all the generations with the same base model into labeled folders. Whew. It was a real learning process but I could have never slogged through so much without help.
A lot of the gallery and showcase images are only the base model and Nova Skyrift V1. The LoRA worked great straight out of the box!
Please see showcase images and gallery for use cases.
SUGGESTED PARAMETERS
The showcase images range from CFG 3 to 7. Accidently I typed "33" instead of "3" for the CFG in some of this generations and got some very cool results. There's plenty of room for more experimenting. These are some notable base models.
SUGGESTED CHECKPOINTS
Realism_Illustrious_By_Stable_Yogi v3.5_FP16
SFW AND NSFW SUPPORT
Nova Skyrift V1 - Future/Fantasy/Gothic Character - SDXL is great for both SFW and NSFW generations. The large majority of training images were PG-13 - R. Please see the use cases for examples of both!
PLEASE LIKE, LOVE, FOLLOW, AND SHARE YOUR GENERATIONS!
Thanks for checking this out! - Timothy