language:
- en
thumbnail: https://drive.google.com/uc?export=view&id=1JWwrxQbr1s5vYpIhPna_p2IG1pE5rNiV
tags:
- text2image
- prompting
license: apache-2.0
datasets:
- succinctly/midjourney-prompts
This is a GPT-2 model fine-tuned on the succinctly/midjourney-prompts dataset, which contains 250k text prompts that users issued to the Midjourney text-to-image service over a month period. For more details on how this dataset was scraped, see Midjourney User Prompts & Generated Images (250k).
While this model can be used to auto-complete prompts for any text-to-image model (including the DALL·E family), note that it occasionally produces Midjourney-specific tags. Users can specify certain requirements via double-dashed parameters (e.g. --ar 16:9 sets the aspect ratio to 16:9, and --no snake asks the model to exclude snakes from the generated image) or set the importance of various entities in the image via explicit weights (e.g. hot dog::1.5 food::-1 is likely to produce the image of an animal instead of a frankfurter).