unbg models ... manages model assets on disk.
Model aliases:
fast=rmbg-1.4quality=rmbg-2.0
RMBG-1.4 and RMBG-2.0 have model-specific conditions, including restrictions or separate permission for commercial use. Installation does not change those terms; review the model license notice before redistribution or production use.
Choose a model profile#
- Use
fastfor lower-resource environments and faster iteration - Use
qualitywhen you want higher-quality segmentation and your hardware can support it
unbg models install#
Installs model assets and prints a JSON report.
unbg models install --model fast
unbg models install --model quality
unbg models install --all
Common options:
--model-dir <dir>: where models are stored--onnx-variant <fp16|fp32|quantized|q8|auto>: ONNX variant to install (default:fp16)--hf-token-env <ENV_NAME>: env var name that contains your Hugging Face token (default:HF_TOKEN)--revision-rmbg14 <rev>(default:main)--revision-rmbg20 <rev>(default:main)--verify-only: validate instead of downloading
unbg models list#
Prints the installed models list (JSON).
unbg models list
unbg models list --model-dir ./models
unbg models verify#
Verifies the model directory and prints details (JSON).
unbg models verify
unbg models verify --model-dir ./models
unbg models update#
Updates one or more models (JSON report). This is similar to install, but intended for refreshes.
unbg models update
unbg models update --model fast
unbg models update --model quality
unbg models update --model all
Running unbg models update without --model updates all supported models.
Options:
--model-dir <dir>--onnx-variant <fp16|fp32|quantized|q8|auto>(default:fp16)--hf-token-env <ENV_NAME>(default:HF_TOKEN)
Related#
- Run inference with installed models: Exec
- End-to-end onboarding path: Quickstart