Nodes/ComfyUI-xiaozhuguang/小珠光图像保存
ComfyUI Node

小珠光图像保存

Save or preview with one node, and stop filling output/ with JPGs you didn't mean to keep

By xiaozhuguang·Created 2 months ago·Updated a day ago· 56
小珠光图像保存
  • images
    output_path
    save_formatJPG
    reduce_lagfalse
    modeSave

    Every ComfyUI user knows the moment: you leave a Preview node in while iterating, forget it's there, and come back to four hundred JPGs clogging output/. 小珠光图像保存 (Xiaozhuguang Image Save) is the save node that bakes a preview/save toggle right into the node - flip it to preview while you iterate, flip it to save when you want a keeper. It's built on ComfyUI's PreviewImage, so the canvas display stays snappy, and it writes JPG (compressed) or PNG (lossless) on demand.

    The nicest touch is lazy encoding: during a run the node only stores the raw pixels (or a fast compressed preview), and the real full-resolution file is encoded when you right-click to save it. So a long batch run doesn't spend its whole time on PNG encoding you may never need. It also accepts a MASK on its images input (it's a wildcard) and saves masks as black-and-white PNGs, which makes it a handy mask-saver too.

    How it works

    The backend takes your images, and in save mode writes them into output/ (or a subfolder via output_path) with a fixed xzg-save_#### filename counter. In preview mode it writes a compressed JPG to temp/ and shows it - nothing permanent, nothing in output/. save_format picks JPG or PNG, reduce_lag trades preview resolution (3840px max side vs 6400) for responsiveness on big batches. The metadata story matches the KB's save-node rules: PNG saves carry the workflow chunks; JPG is a derivative. If you want the A1111-style parameters block that CivitAI auto-links, this node is not that - it's a solid in-pipeline saver, not a share-optimized one.

    The inputs that matter

    • images - IMAGE or MASK (wildcard).
    • mode - Save or Preview. The one you'll flip constantly.
    • save_format - JPG or PNG.
    • output_path - optional subfolder under output/.
    • reduce_lag - lower-res previews for big batches.

    No outputs - it's a terminal node and a cache endpoint.

    Install

    From the ComfyUI-xiaozhuguang pack - Manager (search "ComfyUI-xiaozhuguang"), or:

    cd ComfyUI/custom_nodes/
    git clone https://github.com/xiaozhuguang/ComfyUI-xiaozhuguang.git
    

    Restart. No model downloads.

    Gotchas

    The preview-vs-save footgun is real: the node serializes the mode, and older workflows that stored a Chinese value ("预览") are handled, but if you've hand-edited a workflow and the mode string is anything unrecognized it defaults to Save - which is exactly backwards from what you wanted. And preview files live in temp/, so they vanish on restart. Keep the master as a PNG in save mode; that's the file with the workflow embedded, per the image-I/O rules.

    Categoryxiaozhuguang

    Inputs (5)

    NameTypeDefaultDescription
    images*
    output_pathoptSTRING
    save_formatoptCOMBOJPG2 options: JPG, PNG
    reduce_lagoptBOOLEANfalse
    modeoptCOMBOSave2 options: Save, Preview

    Outputs (0)

    No outputs