保存图像增强版 (Save Image Plus)

ComfyUI's stock Save Image node embeds your entire workflow graph into the output PNG - drag the image back into ComfyUI and the whole pipeline reconstructs, which is genuinely one of the best things about the tool. Save Image Plus does that too, but it also writes a second, older kind of metadata into the same file: A1111-format generation parameters (positive prompt, negative prompt, LoRA syntax, checkpoint name). That's the whole pitch - one save node, two metadata formats, so the image is readable by whatever's looking at it.

Why that second format matters

A lot of tooling outside ComfyUI - prompt viewers, some upload flows, people used to reading A1111-style "parameters" text off a PNG - never learned to read a ComfyUI workflow graph. They know the old flat format: prompt, negative prompt, model name, done. If you only embed the ComfyUI graph, that tooling sees nothing. Save Image Plus writes both, so the same file works for someone who wants to reload your exact graph and for something that just wants to know what prompt and checkpoint produced the image.

The inputs and outputs that matter

Required, and this is most of what you'll actually touch:

• images - what gets saved.
• enable (default on) - flip it off and the node does nothing at all, which is handy for toggling a save step off mid-testing without deleting the node.
• filename_prefix (default ComfyUI) and file_format (PNG / JPEG / WEBP, default PNG) - the usual.
• quality (1–100, default 100) - only matters for JPEG/WebP.
• embed_workflow (default on) - embeds the ComfyUI graph, PNG only. Turn this off if you specifically don't want the graph traveling with the file (sharing an image publicly without exposing your setup, say).
• save_clean_copy (default off) - writes an additional file with no workflow and no metadata baked in at all, for when you want one shareable clean copy alongside the full one.
• enable_preview (default off) - whether the image shows in the UI, independent of whether it's saved to disk.

Optional, and this is where the A1111 metadata comes from - all four are plain strings you can wire in directly:

• positive_prompt, negative_prompt, lora_syntax - feed these from wherever you built your prompt.
• checkpoint_name - manual override, highest priority. If you leave this unwired, pair this node with the pack's own ModelNameExtractor: wire your MODEL into that node and its model_name string output straight into this field, and the saved metadata carries the actual checkpoint filename without you retyping it.

It's an output node - no data flows onward from it.

How to install it

ComfyUI Manager: search "ComfyUI-Danbooru-Gallery", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Aaalice233/ComfyUI-Danbooru-Gallery.git
cd ComfyUI-Danbooru-Gallery
pip install -r requirements.txt

Restart ComfyUI. No models to download - this node only touches files you already generated.

Common issues & troubleshooting

Workflow doesn't embed. embed_workflow only works on PNG. Switch file_format to PNG if you need the drag-back-in behavior; JPEG and WebP can't carry it.

A1111 metadata is missing entirely. All four fields - positive_prompt, negative_prompt, lora_syntax, checkpoint_name - are optional and unwired by default. If nothing's connected, there's nothing to write; this node doesn't reach back into your graph and infer your prompt on its own, you have to pipe it in.

Saved image doesn't match what you expected to see. Check enable first - it's a real switch, not a cosmetic one, and when it's off the node is a no-op even though it's still sitting in your graph looking wired up correctly.

Nothing saves at all / pack doesn't load. Confirm the clone landed at ComfyUI/custom_nodes/ComfyUI-Danbooru-Gallery and that pip install -r requirements.txt finished without errors, then check the ComfyUI startup console for an import failure from this specific pack.