ℹ️ Lora info

You downloaded a LoRA six months ago, and now you can't remember its trigger word. Or you grabbed one off CivitAI and want to know how it was trained - what dim, how many epochs - before you trust it. This node reads that straight out of the .safetensors file. LoRA files can carry metadata baked in at training time: trigger words, tags, training parameters, source URLs. This node pulls it out and hands it to you as text, no digging through JSON by hand.

It's an inspection tool, not a generation one - it doesn't apply the LoRA, it tells you about it. Which is exactly what you want when you're organizing a messy models folder, recovering a lost trigger word, or auditing a download before you use it.

How it works

Kohya-style LoRA trainers (the ones behind most community LoRAs) can embed a metadata block in the safetensors header - the trigger words, the tag frequencies from the dataset, and the hyperparameters the training ran with. This node reads that header and extracts whichever field you ask for. Point it at a file, pick the info you want, get a string back.

The inputs that matter

• lora_path - the path to the LoRA .safetensors file you want to inspect.
• info_type - which piece of metadata to pull. The dropdown has twelve options, and they map to real training concepts:
  • Full_Data - everything, the whole metadata dump.
  • Trigger_word / Tag - the activation words and dataset tags. This is what most people are here for.
  • Url / Copyright - provenance: where it came from, who made it.
  • Dim / Alpha - the LoRA's rank and alpha, which tell you its capacity and how strongly it was scaled during training (a rank-8 style LoRA and a rank-128 character LoRA are very different animals).
  • Unet_lr / Batch_Size / Epochs / Steps - the training recipe, useful if you're trying to reproduce or understand a LoRA's behavior.
  • Info_txt - the free-form info text if one was saved.

The single output is info (STRING) - route it into a text/preview node to actually read it, or into a prompt encoder if you want to auto-inject a trigger word.

Common issues

It returns nothing / blank. The LoRA has no embedded metadata. This is common - plenty of LoRAs on CivitAI were saved without it, or had it stripped, and some trainers just don't write it. No metadata in the file means nothing for this node to read. It's not broken; the data was never there.

Trigger_word is empty but the LoRA definitely has one. Trigger words are only present if the trainer recorded them in the metadata. If it's blank, check the model's CivitAI/HF page - the author usually lists the trigger there even when the file doesn't carry it.

The tags look like a giant frequency list. That's normal - the Tag field is often the dataset's tag counts from training, not a curated trigger list. Use Trigger_word for the activation phrase and Tag to understand what the LoRA was trained on.

Path errors on a hosted instance. lora_path is server-side. On a remote or cloud ComfyUI, the file has to be on that machine, in the LoRA folder it expects.

Installing SDVN_Comfy_node

ComfyUI Manager: search "SDVN_Comfy_node" → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node

Then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from the ComfyUI root and restart. The pack has a companion Model info editor if you want to write metadata back into a file, not just read it. This node is under 📂 SDVN / ℹ️ Info_check.