EPOCH 33
Photorealistic LoRa trained at resolution 1200 on high quality professionally color-graded photo portraits of hikers. Photographs (about 140 pieces) was taken in natural / mountainous setting at very bright sun light.










Gaal'Gai - is a one of the ancient names of the people living in the area where the photographs were taken. This is the series of LoRa trained on thematic photographs from awesome travel in few mountainous regions.
#1
Look into "captions.zip" in order to get an idea of better prompting. Captions in dataset contains two styles of visual description. It is started with natural language. At the end there are WD styled tags. All captions contains one of two triggers of the lens used.
#2
Use split weights of this LoRa to control composition and visual style separately.
Example 1: <lora:gaalgai_hiker_ports_e33:0.67:0.1> - High CLIP (INxx) weight and Low Unet (OUTxx) weight; replicate composition, but avoid affect colors.
Example 2: <lora:gaalgai_hiker_ports_e33:0.1:0.67> - Low CLIP (INxx) weight and High Unet (OUTxx) weight; replicate visual style but avoid affect composition.
This LoRa affect the following aspects:
Background - it produces the ultra realistic natural / mountainous background with bokeh effect.
Subject pose, ???????, skin texture, facial expression - it produces more natural / realistic ??????? (biased toward feminine figures) with realistic natural skin textures. Use ADetailer for best results.
Eyewear, Headdress and body dress - it can produce more natural casual dresses and add sunglasses.
Lighting, colorgrading - it produces more natural light and colors (cooler background hues of mountains and sky; warmer greenery and rocks in foreground).
Two photographic lenses - dataset contains photographs taken with Schneider Super Cinelux 70mm F/2.0 cinematic lens and Tamron 150-500mm F/5.0-6.3 telephoto zoom lens. They both give very different bokeh effects. Schneider gives more smooth and blurry background. Look into "captions.zip" for precise prompts example.
Fantasy characters - during the tests I noticed that this LoRa can produce more natural and realistic hybridization of different characters (see example with woman-arachnid hybrid).
EPOCH 33
Photorealistic LoRa trained at resolution 1200 on high quality professionally color-graded photo portraits of hikers. Photographs (about 140 pieces) was taken in natural / mountainous setting at very bright sun light.
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
