Hi,

created this advanced captioniong workflow and system instructions to generate Captions for Flux for image batches. Since Flux uses two text encoders Clip L (77 tokens) and T5 (256 tokens) I implemented two caption streams. A NL pass for the T5 and a comma seq Pass for Clip L.

This dual approach not only allows for flexible prompting but also maximizes the utility of small datasets by adding two captions per Image. Whether you're aiming for detailed, natural language descriptions or concise, token-efficient sequences, this setup has you covered.

Important tip: If you're diving into this workflow, don't forget to add your OpenAI API key to your root folder for running the GPT-4O component. And if you work with the two caption approach don’t forget to copy the images before starting the caption process

Update: For all open source Hustlers I added Joycap, florence and wd14 tagger into the mix.

Enjoy