Nodes/JPS Custom Nodes for ComfyUI/Revision Settings Pipe (JPS)
ComfyUI Node

Revision Settings Pipe (JPS)

Unpacking the Revision bundle where you actually need it

By JPS-GER·Created 3 years ago·Updated 2 years ago· 100
Revision Settings Pipe (JPS)
  • revision_settings
  • crop_res
  • crop_intpol
  • rev1_crop
  • rev1_offset
  • rev2_crop
  • rev2_offset
  • rev1_strength
  • rev2_strength
  • rev1_noiseaug
  • rev2_noiseaug

This node exists purely to undo what Revision Settings (JPS) just did. That node bundles ten separate values - crop settings, strengths, noise augmentation, for two reference images - into a single BASIC_PIPE wire, which keeps your settings menu tidy but leaves you with a bundle you can't plug into anything. Revision Settings Pipe takes that bundle back apart into its ten individual sockets, so you can drop it right next to the nodes that actually consume each value instead of routing one overloaded wire across your whole canvas.

How it works

It's a straight unpack: one BASIC_PIPE input in, ten typed outputs out. This is the second half of a "menu + pipe" convention JPS runs through most of its Settings nodes - configure everything once in a compact menu node, tucked wherever's convenient, and place the matching Pipe node physically near whatever downstream nodes need the raw values. It's the same underlying idea as rgthree's Context nodes bundling a pile of graph connections down one wire for later unpacking, just scoped specifically to numeric and enum settings rather than general graph objects.

Practically: put this node next to your two CLIP Vision Encode / unCLIP conditioning chains for rev1 and rev2, and each one just grabs the specific outputs it needs.

The inputs and outputs that matter

  • revision_settings (BASIC_PIPE, required) - the bundle from Revision Settings (JPS).
  • Outputs: crop_res (INT), crop_intpol (enum), rev1_crop (enum), rev1_offset (INT), rev2_crop (enum), rev2_offset (INT), rev1_strength (FLOAT), rev2_strength (FLOAT), rev1_noiseaug (FLOAT), rev2_noiseaug (FLOAT).

Wire crop_res and crop_intpol into whatever's cropping your reference images to a CLIP-vision-ready square; rev1_crop/rev1_offset and rev2_crop/rev2_offset into your two crop nodes; and the strength/noiseaug pairs into your unCLIP conditioning nodes for each reference.

How to install it

ComfyUI Manager, searching "JPS Custom Nodes for ComfyUI," or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/JPS-GER/ComfyUI_JPS-Nodes.git

Restart. Nothing extra to install beyond the pack itself.

Common issues & troubleshooting

This node alone doesn't do anything visible. It's pure plumbing - if your Revision workflow isn't producing the effect you expect, the problem is almost never here. Check the settings back on the Revision Settings (JPS) node upstream, or the unCLIP conditioning nodes downstream that actually consume these outputs.

You're using this without knowing what Revision is. Worth a quick gut-check: Revision is Stability's 2023 unCLIP-based image-prompting feature, largely superseded by IP-Adapter for reference-image conditioning. If you're not deliberately chasing the Revision technique, both this node and its settings-menu partner are safe to strip out of a workflow you're simplifying - nothing else in a typical SDXL graph depends on them.

Values look right here but the output looks wrong. Since this node just forwards whatever was set upstream, double-check the actual numbers on the Revision Settings node itself first - if rev1_strength is pinned at 10 or crop_res doesn't match your CLIP vision model's expected input size, that's where to fix it, not here.

CategoryJPS Nodes/Pipes

Inputs (1)

NameTypeDefaultDescription
revision_settingsBASIC_PIPE

Outputs (10)

NameTypeDescription
crop_resINT
crop_intpollanczos,nearest,bilinear,bicubic,area,nearest-exact
rev1_cropcenter,top,bottom,left,right
rev1_offsetINT
rev2_cropcenter,top,bottom,left,right
rev2_offsetINT
rev1_strengthFLOAT
rev2_strengthFLOAT
rev1_noiseaugFLOAT
rev2_noiseaugFLOAT