Giga LoRA Studio — tool models
Every model file the Giga LoRA Studio image tools download, mirrored into one repository so the app pulls from a single known source instead of five third-party accounts.
Nothing here is original work. These are redistributed copies of publicly released weights, unmodified, kept under the filenames the ComfyUI nodes expect on disk. Credit and licensing belong to the original authors, listed below.
What is here, and what uses it
| Folder | Files | Used by |
|---|---|---|
BiRefNet/ |
12 segmentation and matting checkpoints, plus the node's birefnet.py, birefnet_lite.py, BiRefNet_config.py and config.json |
Background removal |
Lama/ |
big-lama.pt |
Watermark removal (the inpainting pass) |
grounding-dino/ |
groundingdino_swint_ogc.pth and its config |
Watermark detection (locating the mark) |
sam/ |
sam_vit_b.pth |
Watermark detection (segmenting what was located) |
upscale_models/ |
RealESRGAN_x2.pth, RealESRGAN_x4plus.pth |
Upscaling |
Folder names match the paths the ComfyUI-RMBG node pack reads from, so a file
can be dropped straight into a ComfyUI models/ tree without renaming.
Sources and licenses
| Model | Upstream | License |
|---|---|---|
| BiRefNet (all variants except Lucida) | 1038lab/BiRefNet, from ZhengPeng7/BiRefNet | Apache-2.0 |
| Lucida | 1038lab/BiRefNet | MIT |
| Big-LaMa | 1038lab/Lama, from advimman/lama | Apache-2.0 |
| GroundingDINO Swin-T | 1038lab/GroundingDINO, from IDEA-Research/GroundingDINO | Apache-2.0 |
| SAM ViT-B | 1038lab/sam, from facebookresearch/segment-anything | Apache-2.0 |
| Real-ESRGAN x2 | ai-forever/Real-ESRGAN | BSD-3-Clause |
| Real-ESRGAN x4plus | xinntao/Real-ESRGAN releases | BSD-3-Clause |
Each upstream project's own license terms apply to its files. If you are an author listed here and would rather this mirror did not exist, open a discussion and it will be removed.