A GPT2 model to generate prompts for SDXL, especially ZavyChromaXL, or similar models.
Trained from the base model. Attach a Style to affect the render further.
Used in text-completion mode, from very short input prompts (A, The, A beautiful, The portrait of [...])
Settings that I use for variety:
max_length: 75 - temperature: 1.1 - top_k: 24 - repetition_penalty: 1.35