D2 Prompt Sanitizer

You know the feeling: you grab a prompt from a Danbooru tag dump, an old A1111 image, or a shared workflow, and it arrives with long_hair, , ,, cascades, commas glued to the next tag, and maybe a LoRA <lora:foo:1> in the middle of it all. D2 Prompt Sanitizer is the tidy-up pass. You drop it between your text source and the prompt encoder, and it normalizes the string without you hand-editing forty tags.

It's a text-formatting node, not a generation node - feed it any string, get a cleaned string out. In the D2 ecosystem it slots right before D2 Prompt or D2 KSampler, but it works fine ahead of a plain CLIPTextEncode too.

What it actually fixes

Five toggles, all on by default, each doing one thing:

• underscore_to_space - long_hair → long hair. This is the big one for tag-based prompts.
• space_after_comma - normalizes whitespace around commas to a single ", " so a , b becomes a, b.
• remove_extra_comma - collapses ,, / , , runs into one comma and strips commas at the start of a line.
• protect_lora - leaves <lora:my_lora:1> untouched. Without this, the underscore conversion would mangle LoRA syntax.
• protect_score - leaves Pony quality tags like score_9 / score_8_up intact. Same reason: those underscores are part of the tag name.

The protection toggles are the part worth understanding. The sanitizer isn't just blindly replacing _ with spaces; it stashes <...> and score_N matches, does its comma/space work, then puts them back. So turning protect_lora off genuinely would corrupt your LoRA call - leave it on unless you know why you don't want it.

The output

One output, prompt - the cleaned string. That's it. Output node it is not; you keep wiring prompt into your encoder or D2 Prompt.

The processing order matters if you're chasing a weird result: underscore conversion happens before comma cleanup, and protected text is restored last. So score_9, won't end up as score 9 , - the score tag gets stashed before any of the comma surgery.

Installing

Part of the D2-nodes-ComfyUI pack - install the pack, not the node.

• ComfyUI Manager → search "D2-nodes-ComfyUI" → Install → restart.
• Or cd ComfyUI/custom_nodes && git clone https://github.com/da2el-ai/D2-nodes-ComfyUI, then restart.

Requirements are just piexif and charset-normalizer, no model files. Watch the pack's one big gotcha: v32.0.0+ requires the newer ComfyUI (V3 schema) - on an older install use a pre-32.0.0 release.

Notes and gotchas

This node won't add commas between tags that never had them - it fixes existing separators, it doesn't invent structure. If your input is 1girl long hair with spaces as separators, the result is 1girl long hair, not 1girl, long hair. Also: don't expect it to dedupe tags; that's a different toggle that lives in a newer version of the pack than the one you'll see listed here. If your prompts keep coming in with duplicate tags, pair the sanitizer with a dedupe pass in your prompt pipeline.

For a string utility node it's genuinely the one I'd reach for - it's deterministic, off-by-default-friendly (every transform can be disabled), and it costs nothing to leave in the graph.