Drop a .safetensors file here
or click to choose
processed in your browser · never uploaded
Drop or click to replace
This file has no __metadata__ block (no training tags or base model recorded).
Summary
- Tensors
- Parameters
- Data types
Training info
Trigger words (top tags)
From the training tag frequency — likely activation tags.
All metadata
Raw header JSON
A tool that opens a downloaded .safetensors LoRA or model in your browser and shows what's inside. It lists the likely trigger words (activation tags), the base model it was trained on, the frequency of training tags, the network dim and alpha, and tensor stats like the tensor count, parameter count and data types. It's for when you grab a LoRA from Civitai or a sharing site and you don't know which tags fire it, you can't tell whether it's built for SD1.5 or SDXL, or the description got lost — the answers are read straight from the __metadata__ embedded in the file itself. The key feature is that it reads only the file header (the JSON of tensor names and training metadata) and never loads the weights, so even a multi-GB checkpoint opens instantly with no waiting. Because trigger words and training settings are your own creative clues and don't need to leave your machine, this tool never uploads the file — all parsing happens locally in your browser.
How to use
- Drop in a .safetensors file or click to choose one (it parses the header only — the weights are never loaded).
- It automatically shows the likely trigger words, base model, training tags, network dim/alpha, and tensor stats.
- Use "Copy all" to paste the activation tags or training info straight into your prompt. The raw header JSON is there too if you want it.
FAQ
Is the file uploaded to a server?
No. All parsing happens in your browser. The .safetensors file is never uploaded, stored, or sent anywhere — it is read only on your device, so it's safe to inspect even an unreleased model or your own training data.
Can it open multi-GB models?
Yes. It never loads the weights — it reads only the header at the start of the file (the JSON of tensor names and training metadata). That's why even a multi-GB checkpoint opens instantly with no waiting.
How are the trigger words determined?
They come from the frequency of the tags recorded during training: the most-used tags are listed as likely activation tags. It's a frequency-based estimate, so double-check against the description on the page you downloaded from.
No metadata shows up — why?
If the file has no __metadata__ block (a LoRA or model that didn't record its training tags or base model), no training info can be shown. You can still see stats like the tensor and parameter counts.
Does it work on regular models (checkpoints), not just LoRAs?
Yes. Any .safetensors file — LoRA, VAE, or checkpoint — has its header read for tensor stats and, when present, the training metadata.
Can it read .ckpt (pickle) files?
No. This tool only handles the safe .safetensors format. .ckpt is a pickle format that can execute arbitrary code, so it isn't supported — convert to safetensors first.