Resolution-robust Large Mask Inpainting with Fourier Convolutions
Paper • 2109.07161 • Published
LaMa Inpainting Model (big-lama)
Mirror of the LaMa (Large Mask Inpainting) model checkpoint, converted to safetensors format for safe, pickle-free loading. Hosted by Æmotion Studio for use with ComfyUI-FFMPEGA.
Model Details
| File | Size | Format |
|---|---|---|
big-lama.safetensors |
~195 MB | safetensors (989 tensors, 51M params) |
Original source: enesmsahin/simple-lama-inpainting
Usage
This model is automatically downloaded by ComfyUI-FFMPEGA when the AI Object Removal skill is used. No manual setup is required if allow_model_downloads is enabled on the FFMPEG Agent node.
Manual Installation
- Download
big-lama.safetensorsfrom this repo - Place it in
~/.cache/torch/hub/checkpoints/
Why safetensors?
The original big-lama.pt is a TorchScript JIT model that uses Python's pickle serialization — which can execute arbitrary code on load. The safetensors format stores only tensor data, eliminating this risk.
Why This Mirror?
This is a first-party mirror maintained by Æmotion Studio to ensure supply chain resilience. If the upstream source becomes unavailable, FFMPEGA will still be able to download the model from this mirror.
License
The LaMa model was originally released under the Apache 2.0 license by Samsung Research.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support