Tensor Blog · model library field guides
Field notes for
real model libraries.
Storage choices are engineering choices: preserve the bytes you need, state what a result does not prove, and recover a separate copy before reclaiming anything.
Field guides
Start with the pressure point.
Every guide leads with the decision you have to make, not a feature list. Tensor Archive appears where a local, exact archive is the appropriate answer.
Stable Diffusion model size: measure the whole library
Use exact dated examples, then account for variants, VAEs, encoders, adapters, caches and outputs before planning disk space.
ComfyUI model backup: keep workflows and reclaim disk space
Inventory native and extra paths, restore one complete workflow family separately, then remove only a proven redundant copy.
Clear the Hugging Face cache safely
Inspect repositories and revisions, let the cache manager understand shared blobs, and verify what remains.
Civitai backup: archive downloaded models locally and verify the restore
A local preservation workflow for files you are entitled to store, without pretending to be a platform mirror.
Free disk space in a LoRA library without changing the weights
Want to save space with LoRAs? Separate archiving, deletion, pruning and quantization before a full SSD forces a bad choice.
SafeTensors storage vs. quantization: can you compress a SafeTensors file?
Why SafeTensors compression and an exact archive answer different technical requirements.
LoRA training checkpoints: practical fine-tuning checkpoint storage
Keep reproducible completed artifacts while being honest about what a storage tool does not manage.
Archive LoRA files as one related family: one base, several adapters
Recognize the repeated-base pattern and test a storage family without promising that every adapter will shrink.
Backup LoRA files with a model-library retention checklist
Separate active runtime data from completed artifacts, then make recovery the completion condition.