Inspect AI model files: Safetensors, GGUF & ComfyUI

Read the metadata inside AI model files without downloading gigabytes of weights. Yellorn decodes the Safetensors / GGUF header and the ComfyUI workflow locally — just the structure, never the tensor data.

Read the header, skip the weights

AI model files are huge — gigabytes of tensor weights. But the part you usually want to inspect is tiny: the header. Yellorn reads only that header, locally in your browser, and never loads the tensor data. So you can check a model's architecture or a workflow's prompt in seconds, without downloading the whole file into the editor.

Drag the file onto the editor to open it.

Editor tab set to Safetensors after dropping a .safetensors file — the buffer shows a computed __summary__ block with tensor_count 5, total_parameters, and a by_dtype breakdown, followed by each tensor's dtype, shape, and byte offsets, with the same structure expanded in the tree view on the right — Drop a .safetensors file and the header decodes instantly — tensor count, parameter total, dtype breakdown, and every tensor's shape — without touching the multi-gigabyte weights.

What each file shows

Safetensors (.safetensors) — the tensor map (each tensor's name, dtype, and shape) plus a computed __summary__. The raw weights are never read.
GGUF (.gguf) — a computed __summary__ (architecture, parameter count, quantization mix, key hyperparameters), the file's full metadata key/value map, and the tensor list.
ComfyUI PNG — the workflow embedded in a generated image: a __summary__ (checkpoint, LoRAs, sampler, size, positive / negative prompts), the API prompt graph, and the UI workflow. The pixels are never decoded.

Useful things to do with it

Compare two checkpoints — open each model's header in a tab and diff the summaries.
Recover a ComfyUI workflow from a PNG someone shared — the API prompt is ready to POST straight to ComfyUI's /prompt endpoint.
Convert the header to JSON for an editable tab, or search it with JMESPath.

Frequently asked questions

How do I view a Safetensors file's metadata without loading the model?

Drop the .safetensors file onto Yellorn. It reads only the file's header — the tensor map with each tensor's name, dtype, and shape, plus a computed summary — locally in your browser. The multi-gigabyte tensor weights are never read or loaded.

What does the GGUF viewer show?

For a .gguf file Yellorn shows a computed summary (architecture, parameter count, quantization mix, and key hyperparameters), the file's full metadata key/value map, and the tensor list — read from the header only, without loading the weights.

Can I extract the workflow or prompt from a ComfyUI PNG?

Yes. Drop the generated PNG and Yellorn reads the workflow embedded in the image: a summary (checkpoint, LoRAs, sampler, size, positive/negative prompts), the API prompt graph ready to POST to ComfyUI's /prompt endpoint, and the UI workflow. The pixels themselves are never decoded.

Does it download the entire model file?

No. Only the header bytes are parsed; the tensor weights are never read into the editor. Inspection happens locally in your browser, so you can check a model's architecture or quantization in seconds without pulling gigabytes of data.

← All help articles · View Parquet, Excel & binary files · What is Parquet? · Decode certificates · Visualize the decode · Usage guide

Where to go next

Try a fix in the editor or browse more articles.

Open editor All articles Usage guide About Yellorn