Files
vslinx-ComfyUI-vslinx-nodes/web/docs/vsLinx_InteractiveDetailer.md
david 4b3b1e5359 feat(detailer): added guided detailer and more (#23)
* feat(detailer): added guided detailer based on impact-pack and lanczos to upscale with model node
* feat(vae-decode): Added batched vae decode + batched tiled vae decode
* feat(bypass-on-bypass): new node that bypasses/mutes on trigger by state
* chore(docs): updated description texts
2026-06-15 22:25:13 +02:00

3.4 KiB

A clone of the Impact-Pack FaceDetailer node without the wildcard field. Instead of writing fragile SEGS-wildcard syntax, the workflow pauses when the detector finds segments and a dialog pops up showing every detected segment (numbered boxes on the full image + a crop preview per segment) with its own prompt textfield. Once you confirm, execution resumes and every segment is detailed with its own positive prompt.

Requires ComfyUI-Impact-Pack. This node is only registered when Impact-Pack is installed; all other vsLinx nodes work without it.

Prompt rules inside the dialog:

  • Empty field → the segment is detailed with the node's base positive conditioning (identical to FaceDetailer without a wildcard).
  • Any text → the text is encoded with the connected clip and replaces the positive conditioning for that segment only. Impact's <lora:name:weight> syntax is supported and applies the LoRA for that segment only.
  • [CONCAT] prefix → the text is concatenated to the base positive conditioning instead of replacing it.
  • [SKIP] → the segment is left completely untouched.

The dialog remembers your last prompts per node (browser localStorage) and prefills them on the next run. Ctrl+Enter confirms. Clicking a box in the overview focuses its textfield. If the page is reloaded while the workflow is waiting, the dialog is restored automatically.

Parameters (in addition to the standard FaceDetailer parameters):

Parameter Type Description
segment_order left-right / top-bottom / largest-first / confidence / detector The order in which segments are numbered in the dialog and processed.
timeout_sec INT How long the workflow waits for the dialog before the on_timeout policy is applied. 0 waits forever. Keep this > 0 if you also run workflows headless via the API, otherwise the queue would hang with no browser connected.
on_timeout use base prompt / skip detailing / cancel run What happens when timeout_sec elapses without an answer.
always_ask BOOLEAN When enabled (default), the node re-executes — and re-asks — on every run, even if all inputs are unchanged. When disabled, unchanged inputs reuse the cached result without showing the dialog.

Outputs:

Parameter Type Description
image IMAGE The detailed image.
cropped_refined IMAGE (list) The refined crop per segment, in dialog order (matching the prompt numbering, unlike Impact-Pack which sorts by size).
cropped_enhanced_alpha IMAGE (list) The refined crops with the segment mask as alpha channel.
mask MASK Combined mask of all detected segments.
cnet_images IMAGE (list) ControlNet preprocessor previews when a SEGS ControlNet wrapper is involved.
used_prompts STRING A newline-joined summary of the prompt used per segment — handy for image-saver metadata.

Notes:

  • The per-segment seed is seed + segment_index, like Impact-Pack.
  • Image batches are not supported (same limitation as Impact's detailers); use image lists instead.
  • Cancelling from the dialog or from the ComfyUI queue cleanly interrupts the run — the server never hangs on a closed dialog thanks to the timeout policy.