ImageToImage Pipe (JPS)

Pairs with ImageToImage Settings (JPS). That node bundles img2img strength, inpainting, and unsampling parameters into one wire; this one unpacks them wherever downstream you actually need the individual values. Standard Settings/Pipe split for this pack, scoped specifically to the img2img direction - no crop sections, no ControlNet preprocessor blocks, just the three things a pure image-to-image or inpainting pass needs.

What comes out of it

One input, img2img_settings - a BASIC_PIPE (JPS's own bundle, a plain tuple, not Impact Pack's object of the same type name). Seven outputs:

• img2img_strength (FLOAT) - feeds a sampler's denoise input for the plain img2img path.
• inpaint_strength (FLOAT) and inpaint_grow_mask (INT) - the inpainting pair, denoise strength and mask-dilation amount.
• unsampler_strength, unsampler_cfg (FLOAT) - strength and CFG for the reverse-diffusion unsampling pass.
• unsampler_sampler and unsampler_scheduler - which sampling algorithm and schedule the unsampling pass uses. Notably a longer, more current list than the equivalent fields in Generation Settings Pipe's output - this pair tracks ComfyUI's newer sampler/scheduler additions more closely.

This node runs no sampling, inpainting, or unsampling itself - it's routing between your settings menu and whichever KSampler, inpaint, or unsampling nodes are actually doing the work elsewhere in your graph.

Why this pair exists alongside Generation Settings

The img2img/inpaint/unsampler block here is a proper subset of what Generation Settings Pipe (JPS) also unpacks - same field names, same concepts - but Generation Settings carries three ControlNet preprocessor sections and a crop section on top of it, meant for workflows that switch between several generation modes without rebuilding the graph. If your workflow is only ever doing img2img or inpainting and never touches Canny, ZoeDepth, MiDaS, or OpenPose, this scoped-down pair is the simpler tool: fewer outputs to wire, fewer places for a connection to go stray, and one obvious node to check if something's misconfigured instead of hunting through thirty possible fields. Worth also noting this pair's sampler and scheduler lists are newer than Generation Settings' baked-in copy of the same fields - 34 sampler options here against 21 there - so even if you're not sure which to reach for, this is generally the more current choice for the unsampling step specifically.

Installing it

Ships with the pack. ComfyUI Manager: search JPS Custom Nodes for ComfyUI, install, restart. Manual:

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

Restart afterward, deleting any earlier copy of the pack first. No models, no extra dependencies.

Where people get tripped up

Without an ImageToImage Settings (JPS) node upstream, this pipe has nothing to unpack. Beyond that, the easiest mistake is only wiring the img2img_strength output and forgetting the unsampling and inpainting outputs exist - if your workflow is meant to use unsampling for a style-transfer-style pass but you've only connected the plain denoise strength, you'll get a normal img2img result instead of the higher-fidelity unsampled one you were expecting. Trace through which of the seven outputs your specific branch actually needs before assuming a value is wrong; it's easy to leave one half of this node's outputs unconnected and not notice until the result looks off.