RMBG-1.4 β€” F16 GGUF

F16 GGUF of BRIA RMBG-1.4 (BriaRMBG / IS-Net), used by the Lux3D TripoSG pipeline for background removal on inputs without a valid alpha channel.

This is an auxiliary model: it is not a Lux3D output family and it is licensed separately from TripoSG.

Model

  • Upstream: briaai/RMBG-1.4
  • Architecture: BriaRMBG (IS-Net); 44.1M parameters.
  • Source: Lux3D canonical model.safetensors, 176,381,952 bytes, 798 tensors (686 F32 parameters + 112 int64 num_batches_tracked buffers).
  • general.architecture = BriaRMBG (top-level config architecture); lux3d.family = rmbg.
  • Output: single-channel foreground mask.

File

file weights size notes
RMBG-1.4-44M-F16.gguf F16 0.08 GiB 686 tensors; 112 rank-0 buffers omitted

F16 is the only quantized build: the network contains no rank-2 linear *.weight matrices, only convolutions and BatchNorm. Integer GGUF dtypes would leave every tensor in the F16 fallback and produce an identical file with no speed or size benefit. The skipped num_batches_tracked tensors are training-only counters and are not used at inference.

Compared with the F32 canonical package (0.16 GiB), the F16 GGUF halves disk usage. End-to-end TripoSG runs with this file produce normal meshes; the matting network is small and F16 is effectively lossless for it.

Usage

lux3d-cli run triposg --device cuda `
  --model-path models/TripoSG-1.5B-Q8_0.gguf `
  --rmbg-model-path models/RMBG-1.4-44M-F16.gguf `
  --source input.png --output out.obj

--rmbg-model-path accepts the .gguf file or a directory containing model.gguf. If a TripoSG package directory is used and no explicit RMBG path is given, the CLI looks for a sibling rmbg-1.4 directory; with a flat GGUF layout, pass --rmbg-model-path explicitly.

Metadata

  • general.architecture = BriaRMBG
  • lux3d.family = rmbg
  • lux3d.quantization = f16, lux3d.fallback_dtype = f16
  • lux3d.resolved_config β€” embedded JSON.

License

Non-commercial. RMBG-1.4 is source-available under the BRIA RMBG 1.4 Model License Agreement; it is not MIT and is not covered by TripoSG's license. See RMBG-1.4-LICENSE.md and https://bria.ai/bria-huggingface-model-license-agreement/.

Commercial use requires a separate agreement with BRIA: https://go.bria.ai/3B4Asxv. Keep RMBG-1.4-LICENSE.md with the file when redistributing. This package grants no rights beyond the upstream license.

Downloads last month
33
GGUF
Model size
44.1M params
Architecture
BriaRMBG
Hardware compatibility
Log In to add your hardware

16-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for oxide-lab/RMBG-1.4-GGUF

Base model

briaai/RMBG-1.4
Quantized
(4)
this model