My first model. Trained to create only cybernetic faces or heads in the style of Adeptus Mechanicus. This creation is far from perfect. The images you receive may lack the usual colors of the color. But I think I was able to achieve the necessary style. I am also open to criticism and would welcome any comments.
Description
Note. Place the model file in the "hypernetworks" folder. It is located on the way (E:\folder where you have located the neural network\stable-diffusion-portable-basic\models\hypernetworks). (Stable Diffusion checkpoint) on which the images were generated was taken from this site from a user under the nickname "xpuct". The trigger for the hypernetworks is (cybernetic hed in a hood <hypernet:Mechanicus1:1>) in a Promt or you can write (cybernetic hed in a hood). After that, click the "Show extra networks" button, which is located under the "Generate" button.
All results were obtained at values such as:
Sampling method: (Euler a) or (DPM++ SDE Karras)
Sampling steps: (from 20 to 30)
Width: <=(832)
Height: <=(896)
CFG Scale: (from 5 to 7)
Seed: -1
Restore faces: the checkmark is removed
Negative promt: disconnected limbs, malformed hands, blurry, watermark, watermarked, oversaturated, censored, distorted hands, amputation, missing hands, obese, doubled face, double hands, neon colors, very bright colors, ((the right face)), ((beautiful face))
FAQ
Comments (8)
01010000 01110010 01100001 01101001 01110011 01100101 00100000 01110100 01101000 01100101 00100000 01001111 01101101 01101110 01101001 01110011 01110011 01101001 01100001 01101000 00100001
01000001 01101110 01100100 00100000 01100100 01101111 01101110 00100111 01110100 00100000 01111001 01101111 01110101 00100000 01100100 01100001 01110010 01100101 00100000 01110011 01110100 01110101 01100100 01111001 00100000 01101000 01100101 01110010 01100101 01110100 01100101 01101011
Beautiful, what is the data?
I think you could double the accuracy if you upscaled your dataset of images beforehand.
Unfortunately, there is no time to do this right now. But the advice is good, it will be necessary to experiment