EmAySee_LoRAMetadataExtractorTagDrop
The same LoRA metadata dump, minus the tag-frequency wall of text
- lora_upload
- metadata_json_string
Identical twin, one toggle different
EmAySee_LoRAMetadataExtractorTagDrop is the metadata extractor you already met, with exactly one extra knob: an exclude_tag_frequency boolean that deletes ss_tag_frequency from the output before it's returned as JSON. That's the entire difference between the two nodes.
Why is that worth a whole separate node? Because ss_tag_frequency is the part of a LoRA's metadata that's enormous and rarely useful - it's a per-image count of every tag the trainer used, and on a properly-trained file it's a nested JSON blob hundreds of lines long that buries the things you actually came for: ss_network_dim, ss_network_alpha, ss_learning_rate, ss_bucket_info, trigger words. If you're inspecting a LoRA to learn its hyperparameters, the tag table is noise. This node lets you say "drop it" and read the signal. If you do want the tag frequency - say, to audit what tags a dataset emphasized - flip the toggle off and you get the full dump, same as the plain extractor.
Inputs
lora_filepath(STRING) - full filesystem path to the.safetensorsLoRA. Absolute paths only; relative names won't resolve. Pair withEmAySee_LoraFilePicker'sfull_filepathoutput.exclude_tag_frequency(BOOLEAN, default off) - the toggle. On = tag table stripped.lora_upload(LORA, optional) - accepted for drag-and-drop compatibility; the code uses the manual path string for actual reading.
Output: metadata_json_string (STRING).
How it works
Everything else is byte-for-byte the parent node's logic: safetensors metadata read, full __metadata__ config parsed as JSON when present, otherwise ss_-prefixed keys extracted with type coercion and nested-JSON parsing for ss_bucket_info/ss_datasets. The only addition is the conditional delete of ss_tag_frequency (it handles both the extracted-keys path and the full-config path, since the key can appear in either place). Errors come back as output text, not crashes.
Install
From ComfyUI_EmAySee_CustomNodes:
cd ComfyUI/custom_nodes
git clone https://github.com/EmAySee/ComfyUI_EmAySee_CustomNodes
# restart ComfyUI
Or ComfyUI Manager → "ComfyUI_EmAySee_CustomNodes". No extra dependencies.
Gotchas
Unlike its sibling, this one sits in a sensible category (EmAySee/LoRA Tools), so search is straightforward. The toggle defaults to off, meaning out of the box you get the same bloated output as the plain extractor - flip it on if that's why you installed this node. And since it's a near-duplicate, don't install it expecting extra features: if you never care about tag tables, this is simply the one of the two you should reach for, and the only reason both exist is that the author published the follow-up rather than editing the original.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| lora_filepath | STRING | — | |
| exclude_tag_frequency | BOOLEAN | false | — |
| lora_uploadopt | LORA | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| metadata_json_string | STRING | — |