A simple workflow implementing Rdancer's <a target="_blank" rel="ugc" href="https://github.com/rdancer/ComfyUI_Florence2SAM2">Florence2+SAM2 automatic masking node</a> to perform automatic, prompt-driven detailing. Arbitrary objects or body parts can be revised in one sweep, using one prompt, separating each target with commas.Note that Florence2 and SAM2 both have to be loaded into RAM (although they can be loaded onto CPU at the obvious cost of speed), so users with RAM-bound setups should use caution.NOTE: The Rdancer node has been defunct for a couple months now; it uses an old version of the Transformers library which has had breaking changes since this workflow was posted. You'll have to install Transformers 4.49.0 manually, by activating the ComfyUI environment and doing <code>pip install transformers==4.49.0</code>. Rolling back hasn't broken anything for me personally, but maybe back things up before you go futzing with the Python packages.

(PARTLY BROKEN, SEE NOTE) Prompt-Driven Detailer (Florence2+Sam2)

I forgot to change a couple of things which led to the SEGS node underperforming. If it's changing more than you expect, uncheck "combined", "bbox_fill" and "contour_fill" in the MASK to SEGS node.

Files with SHA256: 07584c343bb60469d6036431f36162f51bb92875e45d303e8cb263721f09349a