Feature request: Get percentage of how well a generated image match the prompt.

[Patan77]

I would like to request a feature that outputs a floating point value between 0.0 to 1.0 of how well the generated image match the input prompt.

Example Prompt: a photo of a white and grey cat with blue eyes looking at the camera with a serious look on its face, with a pink nose and slightly blue eyes

Then after it have generated the image it evaluate how close the generated image match the given prompt and output a value like "match: 0.7893164"

[grexzen]

Check my repo, I made an aesthetic scorer for this as a script that you can use to score images. It uses the image embeddings, so essentially is doing what you are asking already.