小珠光图像保存-化神级

ComfyUI's stock Save Image has one setting: a filename prefix that always lands in the output folder root. That's fine until you run a hundred jobs and the output folder is a wall of xzg-save_00001.jpg files. The 化神级 ("transcendent-grade" - the pack author names their nodes like power levels in a cultivation novel) ImageSaveCustom is the version of that node that actually lets you organize what you save. It's part of the Xiaozhuguang pack, a Chinese-first ComfyUI utility suite, so don't be surprised that the node shows up in Chinese ("小珠光图像保存-化神级") until you switch the UI to English.

It's a save node, so it does exactly one thing at the end of your pipeline: write images to disk. What makes it worth reaching for over the built-in is the custom output path. Flip use_default_output off and base_dir becomes a real path - absolute, or relative to ComfyUI/output/ - and it supports template variables: {date}, {time}, {datetime}, {workflow}, {node_id}, {format}. So renders/{date}/{workflow}/ turns into renders/2026-08-25/my_workflow/ automatically, which is the difference between a scrapheap and an archive.

The inputs that matter, in order:

• images - the batch you want saved (the only required input).
• use_default_output - on (default) behaves exactly like the pack's plain save node and ignores everything else; off unlocks the custom path.
• base_dir - the folder, with the {date}/{workflow} style templates above.
• filename_custom - prefix for the file (default xzg-save).
• add_date_stamp / add_time_stamp - append date or time to filenames so you don't overwrite on re-runs.
• save_format - JPG (compressed, default) or PNG (lossless).
• mode - Save or Preview. Preview shows you the image without writing it. Don't ignore this: the pack fixed a nasty bug where typing a Chinese "预览" into mode still wrote the file anyway (v11.5.8).
• reduce_lag - caps the saved side at 3840px and quality at 85 for big batches. You'll feel it on slow machines.

There are no outputs - it's a terminal node.

Preview is always a compressed JPG for smoothness, whatever you picked, so check the file on disk rather than trusting the on-canvas thumbnail for PNG quality checks.

Install is the standard pack story: ComfyUI Manager → search ComfyUI-xiaozhuguang (or just "xiaozhuguang") → install → restart, or clone it by hand:

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

Then restart ComfyUI. This node needs no model downloads and no extra pip packages - requirements.txt only pins imageio-ffmpeg for the video side, so the save node works out of the box. The only real gotcha: the pack's other, heavier nodes (Qwen, TTS) pull in big dependencies when Manager installs the registry version, so on a lean setup install it from git instead and you keep the whole pack light.

If your files land in the wrong place, you're almost certainly on an old version - path resolution for relative base_dir values was genuinely broken before v11.5.8 and silently dumped everything in the output root. Update and retest.