Like a week ago I finished some new workflows. One uses a LLM to generate a natural language prompt, which is then converted to booru tags via TIPO (rather badly, but you can help it by adding a few relevant tags you want to hold constant).
The other one uses an LLM to auto fill in the speech bubbles based on the description or prompt itself. I guess you could use a VLM for this but idk if they're any good or if they can do lewd. It can handle multiple speech bubbles but it generates each bubble independently so for some prompts it can be repetitive for short texts.
Also sorry the workflow is literally everything I've ever done. But it's all in groups so you can just check the muters. Also the img2colors requires you to install requirements.txt via comfyui manager and then also replace image.view with image.reshape. However that node is only used to match the new text with the original AI text colors, you can bypass it & set colors manually.
https://files.catbox.moe/sf8d84.pnghttps://files.catbox.moe/rgukno.pnghttps://files.catbox.moe/oqx5hu.png