What's this?

• This is a workflow that replaces sensitive images with the well-known distasteful dog image
• By the way, we are using LLM, but there is hardly any reason to use LLM

How?

• Generate tags for the image using JoyTag
• Determine whether there are any sensitive tags among them using merlinite-7b
• If there are, replace the image with a dog image

Punchline of the story

• Most people probably noticed, but there’s really no need to use LLM.
• All you have to do is create a tag set with sensitive tags and compare it with what JoyTag generated.
• You might be able to use it to identify images other than sensitive ones if you change the prompt…
• If you want to waste computational resources, by all means, give it a try😏