V2: Smaller file size. Trained on more than 2x the images for 2x the steps per image. More details at 0.8-1.0 strength, can be turned down to 0.3 to follow prompts better and get more image variety.
V1: This is one of my first training experiments. The data set consists of just 18 of my gnarliest monster gens, trained for 1080 steps in 23 minutes on my 3060. No captioning, the trigger word is just "demonic aliens". I think it turned out surprisingly good for a first try. I trained on sd 1.5 base at 768x768, so it should work with most checkpoints.
Example images used EpicPhotogasm X, DPM++ 2M Karras @ 25 steps and 7 cfg, hi-res fix is 2x latent @ 0.50 denoise.
Prompts were very simple, like:
"a photo of demonic aliens visiting times square, 4k, cinematic lighting, intricately detailed"
"a photo of demonic aliens playing billiards, 4k, cinematic lighting, intricately detailed"
"a photo of demonic aliens playing poker, 4k, cinematic lighting, intricately detailed"
"a photo of demonic aliens frolicking in a field of sunflowers, 4k, cinematic lighting, intricately detailed"
"a photo of a cute kitten, 4k, cinematic lighting, intricately detailed, demonic aliens"
"a photo of a wild horse, 4k, cinematic lighting, intricately detailed, demonic aliens"
"a macro photo of a beautiful flower, 4k, cinematic lighting, intricately detailed, demonic aliens"
Negative prompt was just:
"text, watermark, blurry, ????, ?????"
Weight of 0.8-1.0 worked well.
Anyway, I hope you have fun with this one. :)
If there is enough interest, I might try to make a better version, or maybe multiple flavors of creature horror LoRas.
Let me know if you like it, I'd love to see what you make with it as well!
V2: Smaller file size. Trained on more than 2x the images for 2x the steps per image. More details at 0.8-1.0 strength, can be turned down to 0.3 to follow prompts better and get more image variety.
1. 재게시된 모델의 권리는 원 제작자에게 있습니다.
2. 모델 원작자가 모델을 인증받으려면 공식 채널을 통해 SeaArt.AI 직원에게 문의하세요. 저희는 모든 창작자의 권리를 보호하기 위해 노력합니다. 인증하러 이동
