Sillytavern Expressions Workflow

Another ST workflow, but hopefully not as big as some of the others I've seen. On my 3060, images take 15s to finish, and all 28 takes about 7 minutes. Mask bits partially stolen from this.

IMPORTANT: Set batch size to 28!!! lol

I've screwed up this WF a few times cos of a few misclicks, so be careful.

git clone https://github.com/audioscavenger/save-image-extended-comfyui/ into your Custom Node directory, the Manager's version of the node is outdated. If you use the old version, "-1" will be added to all the images, and you'll have to get rid of them manually. Get other missing nodes as normal.

Things (By group)

Lines: Change "Emote2" or "Main tags" as much as you like. Emote1 is the file name, so don't mess with it.

Images+Prompts: Choose a model, add the name, and put in an "Extra" prompt for additional things like personality traits.

Masquerade: "Florance = True" inpaints the entire image when it's disabled, which gives different poses. Often screws up clothes, so experiment with it and see what happens. Make "Dilate Mask" higher if the mask isn't big enough. If you haven't used it before and want to change the notification sound, put another audio file where notify.mp3 is in your Comfy folder, then change "notify.mp3" to its filename.

Nothing else to change, the KSampler and Detailer have to be 8 steps and 1 cfg for the LORA.
If you want to see slightly more of how the WF works, that "Florence2Run" node is a bit dodgy. Didn't hear it from me, though!!

Nodes aligned with NodeAligner:
https://github.com/Tenney95/ComfyUI-NodeAligner