🦊 RS Save Image LoRA

If you batch-test LoRAs, you've lived this: twenty outputs, all named ComfyUI_00001 through ComfyUI_00020, and you cannot tell which one used which LoRA without opening each file. RS Save Image LoRA is a drop-in replacement for SaveImage that appends the LoRA names and strengths to every filename automatically. Your output becomes hero_style_flux1dev_v0.6_00001.png instead of ComfyUI_00001.png. Files that name themselves - and they never guess wrong, because it reads the actual generation, not your memory.

How it works

Stock SaveImage doesn't know or care what LoRAs ran - it just saves pixels. This node reads the execution payload being sent to the ComfyUI server and collects every LoRA participating in the current generation, then bakes them into the filename. That's why it works with any LoRA loader: it doesn't parse your workflow for a specific node, it reads what actually executed.

Details that matter:

• Chain order preserved - multiple LoRAs are listed in the order they apply, separated by ; .
• Disabled slots skipped - a LoRA toggled off in its loader never appears in the name.
• Clean names - subfolder paths (QWEN\my_lora.safetensors) and extensions are stripped, so you get my_lora_v0.8, not a path blob.
• Everything else is still SaveImage. Full prompt and workflow metadata get embedded in the PNG, so dragging the file back into ComfyUI restores the graph. A / in the prefix still makes subfolders. %batch_num% still works for batches.
• Architecture-agnostic - it reads the workflow, not the model, so it works identically on SD 1.5, SDXL, Flux, Qwen Image, Z-Image, Wan, whatever.

Inputs are exactly the two you'd expect: images and filename_prefix (default ComfyUI). There are no outputs - it's a terminal save node, just like the original.

When it's a must-have

Pair it with RS LoRA Tester in batch mode and the workflow becomes self-documenting. Twenty LoRAs, one queue press, and every output is labeled with exactly what produced it. Same story if you're dialing in strength values: a 0.5 / 0.7 / 0.9 comparison run lands on disk with the strengths in the names.

Even outside batch testing it's quietly useful. Three months from now, when you're staring at an old render wondering what LoRA created that style, the answer is in the filename - no metadata forensics required.

Common issues

• The suffix doesn't appear - if your generation genuinely used no LoRAs, there's nothing to append; you just get a normal save. If you did use LoRAs and the name is still bare, check that the loader actually enabled them (disabled slots are skipped by design).
• Filename length - a big LoRA stack plus strengths makes long names. That's the price of self-documentation; trim your prefix if it bothers you.
• Metadata preserved - yes, the PNG still carries the workflow, so sharing a labeled image doesn't lose the graph.

Installing it

Part of the RaykoStudio pack. ComfyUI Manager → search ComfyUI_RaykoStudio → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Raykosan/ComfyUI_RaykoStudio

Restart ComfyUI. Dependencies (pycairo, opencv-python, scipy, Pillow>=10) install via Manager; a manual clone needs pip install -r requirements.txt. Apache 2.0, tested on ComfyUI 0.15+, Python 3.10+, Torch 2.8+.