Sign In

'Alignment' turn humans to robots (concept replacement example)

Download

1 variant available

fp16 SafeTensor

release_alignment_2000_steps-fp16-no-ema.safetensors

Half precision, best balance • 1.99 GB

Verified:

Type
Checkpoint Trained
Stats

417

Reviews
Published

Mar 29, 2023

Base Model

SD 1.5

Training
Steps: 2,000
Epochs: 1
Hash
AutoV2
1A58400F79
Trigger Words
human
default creator card background decoration
Followers - 2928

2.9K

Likes - 37221

37.2K

Supporter Badge May 2023

Use the trigger word 'human'.

'Alignment' is an example of using conceptmod to replace the 'human' concept with a 'robot' concept. It has been trained for 2000 steps. It sometimes replaces 'woman'/'man' without the 'human' trigger, but not always.

Trained using:
https://github.com/ntc-ai/conceptmod

Training does not require a dataset, only words.

Beware merging this mod. Weights are changed significantly as all humans are now machines.

Trained on:

"#:0.4|human=robot:0.8|robot%human:-0.1"

  • '#:0.4' means resist changing the empty (unconditional) prompt ''.

  • 'human=robot:0.8' - pull humans to robots

  • 'robot%human:-0.1' - make robots and humans closer together based on orthogonality.

  • Technique based on https://github.com/rohitgandikota/erasing

  • Model based on criarcys-fantasy-to-experience