This new technique will surely come in handy to many people, the images in the dataset are all solid red .PNG images.
The Filenames are a selection of Prompts from Lexica.art
Trigger word is Case Sensitive: "Cyber-Samurai"
This demonstrates the concept that we can choose "good prompts" & "Inverted Text" without use of any Image data whatsoever.
note: this technique requires a prompt, using the trigger alone will produce wildly abstract results as the AI needs an anchor in the latent space to allow the elements from the prompt selection to resurface.
MODELS MATTER:
Images below show the outputs using the same prompt & seed with the default released 2.1-768 (top)
versus my own fine tuned 2.1-768 model (bottom)