Adapter folder
~/.tapioca/adaptersKeep the weights you already have
Bring an existing LoRA into Tapioca, copy a complete adapter library to another computer, or reuse a compatible catalog model you already downloaded.
Existing download
Use this when a browser, Civitai client, Hugging Face tool, or another application already downloaded a LoRA to your computer.
tapioca adapter import ~/Downloads/cinematic-motion.safetensors --base minimax-h3 --name cinematic-motiontapioca adapter import "C:\Users\Carlos\Downloads\cinematic-motion.safetensors" `
--base minimax-h3 `
--name cinematic-motion--base is the model family named on the LoRA's model card. It is required because a .safetensors extension does not prove compatibility. --name gives the local import an easy reusable name.
Tapioca verifies the safetensors header, calculates a SHA-256 hash, copies the file into its managed library, and prints a local:// reference. Your original file is left unchanged.
tapioca adapter listtapioca video minimax-h3 --adapter 'local://cinematic-motion#cinematic-motion.safetensors@0.8' --prompt "A cinematic tracking shot" --output adapted.mp4adapter import PATH accepts a file already on your computer. --file selects one weight inside a provider repository; it is not a local path option.
No terminal needed
.safetensors LoRA.Tapioca's desktop and CLI use the same managed adapter library. An import made in either interface appears in the other.
Computer to computer
This is faster than importing files one at a time and preserves provider references, hashes, revisions, and compatibility hints.
~/.tapioca/adapters%USERPROFILE%\.tapioca\adaptersadapters directory from the source.snapshot.json next to its weight files.tapioca adapter list on the destination.If you only have a raw .safetensors file, use adapter import. If you copy Tapioca's managed library, copy each whole folder including snapshot.json.
Large base weights
Base-model variants are hardware specific. Transfer only a model compatible with the destination. An Apple Silicon MLX model is not a Windows CUDA model.
tapioca list on the source and note the exact model ID.TAPIOCA_HOME/models.TAPIOCA_HOME/models with the same directory name.tapioca pull MODEL on the destination without --force.tapioca pull minimax-h3
tapioca listTapioca reuses every matching file, downloads only required files that are missing, and registers the model with paths that are correct for the destination computer.
The registry contains absolute paths from the old machine. Let tapioca pull MODEL rebuild model registration after the files are copied.
Storage control
Set TAPIOCA_HOME before importing or pulling. Models, adapters, voices, recipes, and managed runtimes will use that location.
export TAPIOCA_HOME="/Volumes/ExternalSSD/tapioca"
tapioca adapter list
tapioca list$env:TAPIOCA_HOME = "D:\Tapioca"
tapioca adapter list
tapioca listAdd the environment variable permanently in your operating system if every session should use the external disk. Leave space for partial downloads and generated media as well as final weights.
Before cleanup
tapioca adapter list
tapioca list
tapioca run MODELFor an image or video model, create one small smoke-test output. Delete the old copy only after the destination loads the model and produces a valid result.

Import raw LoRAs, preserve managed metadata, and let Tapioca verify transferred base-model folders.
Continue to the beginner guide →