|
--- |
|
license: creativeml-openrail-m |
|
datasets: |
|
- Nacholmo/controlnet-closesttoblack |
|
--- |
|
|
|
# Controlnet model for use in qr codes |
|
|
|
Conditioning only 15% of the pixels closest to black, so as not to affect the luminance of the rest of the image. ̶I̶t̶ ̶g̶o̶e̶s̶ ̶1̶5̶0̶0̶ ̶s̶t̶e̶p̶s̶ ̶s̶o̶ ̶f̶a̶r̶ ̶a̶n̶d̶ ̶i̶s̶ ̶q̶u̶i̶t̶e̶ ̶p̶r̶o̶m̶i̶s̶i̶n̶g̶ |
|
|
|
### The concept seems to work, now I have to improve the dataset and do the training again, this week I think I will. Also I want to make a Preprocessor to have a blur slider |
|
|
|
1500.ckpt is the automatic1111 controlnet extension compatible weight, 1500.yaml is also important. |
|
|
|
|
|
## 1.5k steps cherry picked examples (not scannable yet) |
|
Steps: 50, |
|
Sampler: DPM++ 2M Karras, |
|
ControlNet: |
|
preprocessor: none, |
|
model: 1500 [dafc5f57], |
|
weight: 1.5, |
|
starting/ending: (0, 0.9), |
|
controlNet is more important |
|
 |
|
 |
|
 |
|
 |
|
|
|
|
|
# New weights up to 7k steps |
|
|
