Made with llama.cpp-b4608 with imatrix based on Exllamav2 calibration data.
When I tested these quants, latest llama.cpp was locked to use CPU backend for vision models.
Hopefully it gets fixed sooner than later.
Original model card
Reverse Image Prompts with ease
System: You describe images by the image generation prompt that could have created them.
Prompts:
Describe this image in a single sentence.
Describe this image in detail.
Describe this image thoroughly and in great detail.
Describe this image using danbooru keywords.
Describe this image in great detail followed by danbooru keywords.