Verified: a year ago
SafeTensor
Type | |
Stats | 934 5,290 |
Reviews | (155) |
Published | Nov 7, 2023 |
Base Model | |
Training | Steps: 6,600 Epochs: 4 |
Trigger Words | binary code and digital art binary code digital art only zero and one |
Hash | AutoV2 80D7FE48C5 |
Please do not share it on other websites at will. I will upload it myself if possible.At least ask for my own opinion first.
After testing, it seems that there is an overtraining situation,still have many issues that need to be adjusted, but they won't be updated these day.
Weight:0.7~1.0
Trigger prompts: binary code, digital art, only zero and one.
Optional prompts: tree, city, room,
car, roadster car (there are issues with training images, and sometimes the side mirrors separate from the cars).
(I didn't realize that someone had already made a similar style until the training was completed:https://civitai.com/models/152610/hacked-tech-world-morph)
Trained it with 55 images for 6600 steps.
This included a substantial amount of images generated by AI from Bing and a small number of search results from the internet.
Originally, I wanted to train a special effect of binary code (numbers 0 and 1) floating in the air as a separate Lora.
However, the majority of images generated through AI couldn't fully meet this condition, but I really liked the style.
Have fun.
If you find that you can't generate any numbers no matter what, you can try to increase the weight of Lora or related prompt words, or check if there are text, number and other prompts in the negative prompts and remove them.